{ "name": "AI Disruption Forecast Impact Register", "version": "1.1", "lastReviewed": "2026-10-06", "methodologyUrl": "https://aidisruptionforecast.org/what-it-means", "recordCount": 67, "areas": [ { "name": "Work & income", "summary": "AI is already improving some bounded tasks, especially for less-experienced workers, while evidence also shows displacement and slowdowns in exposed work. The near-term question is how the gains and losses are distributed, not whether every job moves in the same direction.", "counts": { "total": 9, "Gain": 4, "Harm": 5, "Mixed": 0, "Confirmed": 3, "Partial": 4, "Watch": 2 } }, { "name": "Health & medicine", "summary": "AI can improve screening and access, but poorly validated tools and automated decision systems can also miss patients or constrain care. The strongest benefits still depend on clinical validation, careful deployment, and meaningful human oversight.", "counts": { "total": 8, "Gain": 3, "Harm": 5, "Mixed": 0, "Confirmed": 1, "Partial": 4, "Watch": 3 } }, { "name": "Education & skills", "summary": "Structured, supervised uses can save teacher time and improve learning. Unguarded use can weaken unassisted performance, widen access gaps, or wrongly flag student work.", "counts": { "total": 11, "Gain": 6, "Harm": 5, "Mixed": 0, "Confirmed": 2, "Partial": 6, "Watch": 3 } }, { "name": "Cost of living & productivity", "summary": "AI is creating substantial consumer value and may raise firm productivity. Data-center demand, personalized pricing, and uneven adoption could also increase household costs or leave smaller firms behind.", "counts": { "total": 9, "Gain": 3, "Harm": 6, "Mixed": 0, "Confirmed": 0, "Partial": 2, "Watch": 7 } }, { "name": "Independence & accessibility", "summary": "AI can restore communication and expand access to information and navigation. Reliability, bias, privacy, and false accessibility claims determine whether those tools create genuine independence.", "counts": { "total": 10, "Gain": 5, "Harm": 4, "Mixed": 1, "Confirmed": 1, "Partial": 7, "Watch": 2 } }, { "name": "Safety, fraud & privacy", "summary": "The same technology can detect fraud, reduce crashes, and strengthen cyber defense while also scaling scams, abuse imagery, discriminatory identification, and phishing.", "counts": { "total": 12, "Gain": 5, "Harm": 7, "Mixed": 0, "Confirmed": 1, "Partial": 9, "Watch": 2 } }, { "name": "Science & discovery", "summary": "AI is already changing weather forecasting, biology, chemistry, mathematics, and experimental workflows. Most gains remain domain-specific, and claims of novelty or accuracy still require independent checks.", "counts": { "total": 8, "Gain": 6, "Harm": 2, "Mixed": 0, "Confirmed": 0, "Partial": 8, "Watch": 0 } } ], "records": [ { "id": "I-001", "area": "Work & income", "direction": "Gain", "claim": "In a staggered production rollout at one large firm, access to a generative AI assistant raised customer-support agents' issues resolved per hour by about 15% on average, with the largest gains for less-experienced agents.", "status": "Confirmed", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Multi-country", "whoGains": "Less-experienced and lower-skilled customer-support agents; the deploying firm", "whoBearsTheCost": "Analysis: most-experienced agents saw small quality declines in the study; entry-level support hiring may fall if productivity gains substitute for headcount (not sourced)", "resolutionCriteria": "True if by 2028-12-31 at least one additional independent study reports a support-agent productivity gain of 10% or more; False if independent studies find no significant gain; Partially true if gains are found but below 10% or only for novices.", "ledgerEntry": "By 2028-12-31, gen-AI assistance for customer-support agents will show independently replicated productivity gains of 10% or more at Pilot reach or above in at least one additional firm.", "sources": [ { "id": "SRC-R001", "role": "Supporting source", "title": "Generative AI at Work (Quarterly Journal of Economics, 2025)", "authorOrOrganization": "Brynjolfsson, Li, Raymond", "publicationDate": "2025-05-01", "url": "https://digitaleconomy.stanford.edu/news/generative-ai-at-work/", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Staggered rollout across 5,172 agents; AI access raised issues resolved per hour about 15% on average, most for less-experienced agents; small quality declines for top agents.", "limitations": "One firm, one tool; unnamed deployer supplied data. Link is the authors' lab page; DOI 10.1093/qje/qjae044. Publication month approximate." }, { "id": "SRC-001", "role": "Supporting source", "title": "Generative AI at Work", "authorOrOrganization": "National Bureau of Economic Research", "publicationDate": "2023-04-24", "url": "https://www.nber.org/papers/w31161", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Among 5,179 support agents, AI assistance raised issues resolved per hour by 14% on average and 34% for novice and lower-skill workers.", "limitations": "Observational rollout at one firm; task and workforce may not generalize." }, { "id": "SRC-049", "role": "Supporting source", "title": "Generative AI at Work", "authorOrOrganization": "National Bureau of Economic Research", "publicationDate": "2023-04-24", "url": "https://www.nber.org/papers/w31161", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Among 5,179 support agents, AI assistance raised issues resolved per hour by 14% on average and 34% for novice and lower-skill workers.", "limitations": "Observational rollout at one firm; task and workforce may not generalize." }, { "id": "SRC-085", "role": "Counterevidence", "title": "The effects of generative AI on productivity, innovation and entrepreneurship", "authorOrOrganization": "OECD", "publicationDate": "2025-11-01", "url": "https://www.oecd.org/en/publications/the-effects-of-generative-ai-on-productivity-innovation-and-entrepreneurship_b21df222-en.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Review finds task-level gains are heterogeneous and frequently fail to translate directly to firm or economy-wide outcomes.", "limitations": "Synthesis combines unlike tasks, tools and study designs." }, { "id": "SRC-R080", "role": "Counterevidence", "title": "Large Language Models, Small Labor Market Effects (NBER WP 33777)", "authorOrOrganization": "Humlum, Vestergaard", "publicationDate": "2025-05-09", "url": "https://ideas.repec.org/p/nbr/nberwo/33777.html", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Linking chatbot adoption surveys (25,000 workers, 7,000 workplaces, 11 exposed occupations) to administrative records, difference-in-differences finds precise null effects on earnings and hours, ruling out effects above 1 to 2%; time savings about 3%.", "limitations": "Denmark only; 2023 to 2024 tools; working paper." }, { "id": "SRC-R099", "role": "Counterevidence", "title": "Large Language Models, Small Labor Market Effects (NBER WP 33777)", "authorOrOrganization": "Humlum, Vestergaard", "publicationDate": "2025-05-09", "url": "https://ideas.repec.org/p/nbr/nberwo/33777.html", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Linking chatbot adoption surveys (25,000 workers, 7,000 workplaces, 11 exposed occupations) to administrative records, difference-in-differences finds precise null effects on earnings and hours, ruling out effects above 1 to 2%; time savings about 3%.", "limitations": "Denmark only; 2023 to 2024 tools; working paper." } ] }, { "id": "I-002", "area": "Work & income", "direction": "Harm", "claim": "Evidence suggests early-career employment in highly AI-exposed US occupations was about 19% below the less-exposed counterfactual in the August 12, 2026 study revision using data through June 2026. This is an association; AI’s causal contribution is not isolated.", "status": "Partial", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O4", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "More-experienced workers (35 and older) in the same exposed occupations, whose employment was stable or grew", "whoBearsTheCost": "Early-career workers aged 22 to 25 in software development, customer service and clerical occupations", "resolutionCriteria": "True if by 2028-12-31 a peer-reviewed or second independent administrative-data study confirms a relative employment decline of 10% or more for early-career workers in high-exposure occupations; False if independent analyses find no relative decline; Partially true if a decline is confirmed but AI attribution is rejected or not supported.", "ledgerEntry": "By 2028-12-31, early-career employment in the most AI-exposed US occupations will remain at least 10% below the relative trend of less-exposed occupations in independent administrative data.", "sources": [ { "id": "SRC-R002", "role": "Supporting source", "title": "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence (Nov 2025 version)", "authorOrOrganization": "Brynjolfsson, Chandar, Chen (Stanford Digital Economy Lab)", "publicationDate": "2025-11-13", "url": "https://digitaleconomy.stanford.edu/news/canaries-in-the-coal-mine/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Workers aged 22 to 25 in most AI-exposed occupations saw a 16% relative employment decline after controlling for firm shocks; older workers stable or growing; effects concentrated where AI automates.", "limitations": "Working paper, not peer-reviewed; proprietary ADP data. Earlier version reported 13%." }, { "id": "SRC-R003", "role": "Supporting source", "title": "Summary of Canaries in the Coal Mine sample", "authorOrOrganization": "Silicon Canals", "publicationDate": null, "url": "https://siliconcanals.com/?p=104411", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Main sample tracks 3.5 to 5 million ADP workers per month, Jan 2021 to Sep 2025.", "limitations": "Secondary summary; publication date not captured." }, { "id": "SRC-R072", "role": "Supporting source", "title": "The Labor Market for Recent College Graduates", "authorOrOrganization": "Federal Reserve Bank of New York", "publicationDate": null, "url": "https://www.newyorkfed.org/research/college-labor-market?os=wi", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Recent-graduate unemployment about 5.6% through Q2 2026.", "limitations": "Not AI-specific." }, { "id": "SRC-086", "role": "Counterevidence", "title": "Tracking the Impact of AI on the Labor Market", "authorOrOrganization": "Yale Budget Lab", "publicationDate": "2026-04-16", "url": "https://budgetlab.yale.edu/research/tracking-impact-ai-labor-market", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Updated analysis finds occupational mix, employment and unemployment have not yet changed in a way clearly connected to AI exposure.", "limitations": "Aggregate analysis may miss localized and age-specific effects." }, { "id": "R66-E22-1", "role": "Supporting source", "title": "Revised Canaries employment study", "authorOrOrganization": "Stanford Digital Economy Lab", "publicationDate": "2026-08-12", "url": "https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Early-career employment in exposed occupations is 19% below the less-exposed counterfactual. Decline is mainly reduced hiring. No economy-wide displacement established.", "limitations": "Descriptive association, not a causal estimate; the revised analysis does not isolate AI’s causal contribution." }, { "id": "R66-E23-1", "role": "Counterevidence", "title": "NY Fed business AI survey", "authorOrOrganization": "Federal Reserve Bank of New York", "publicationDate": "2026-09-01", "url": "https://libertystreeteconomics.newyorkfed.org/2026/09/businesses-are-using-ai-to-transform-work-not-cut-jobs/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Regional business survey describes task reorganization and limited current AI-related job reductions.", "limitations": "Employer reports and regional coverage do not invalidate localized early-career payroll effects." } ], "lastReviewed": "2026-10-06" }, { "id": "I-003", "area": "Work & income", "direction": "Harm", "claim": "On a leading global freelancing platform, job posts for automation-prone writing and coding work fell 21% within eight months of ChatGPT's release relative to manual-intensive work, and image-creation posts fell 17% after image generators launched.", "status": "Confirmed", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Multi-country", "whoGains": "Clients substituting AI for freelance work; freelancers winning the remaining, more complex and higher-paid automation-prone jobs", "whoBearsTheCost": "Online freelancers in writing, coding and image creation, who faced fewer posts and more competition per post", "resolutionCriteria": "True if by 2028-12-31 at least one independent study on another platform or later period confirms a decline of 15% or more; False if independent evidence shows recovery to trend; Partially true if declines are confirmed but smaller than 15%.", "ledgerEntry": "By 2028-12-31, demand for automation-prone freelance writing, coding and image work will remain at least 15% below pre-ChatGPT trend on at least one additional major platform.", "sources": [ { "id": "SRC-R004", "role": "Supporting source", "title": "Who Is AI Replacing? The Impact of Generative AI on Online Freelancing Platforms (Management Science 71(10))", "authorOrOrganization": "Demirci, Hannane, Zhu", "publicationDate": "2025-10-01", "url": "https://spiral.imperial.ac.uk/entities/publication/88113a91-b994-4717-8526-dadac657f1cb", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "21% relative decline in automation-prone writing and coding job posts within eight months of ChatGPT; 17% decline in image-creation posts after image generators; remaining jobs more complex and better paid.", "limitations": "One platform; short post-period." }, { "id": "SRC-R005", "role": "Supporting source", "title": "Generative AI reduces demand for some freelance jobs", "authorOrOrganization": "The Decoder", "publicationDate": null, "url": "https://the-decoder.com/generative-ai-reduces-demand-for-some-freelance-jobs-in-writing-coding-and-design-study-says/", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Study covered about 1.4 million job listings, July 2021 to July 2023.", "limitations": "Secondary; other summaries cite about 2 million posts across 61 countries. Publication date not captured." }, { "id": "SRC-007", "role": "Supporting source", "title": "Is generative AI a job killer? Evidence from the freelance market", "authorOrOrganization": "Brookings Institution", "publicationDate": "2024-09-10", "url": "https://www.brookings.edu/articles/is-generative-ai-a-job-killer-evidence-from-the-freelance-market/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "More-exposed freelancers experienced about 2% fewer contracts and 5% lower earnings after ChatGPT.", "limitations": "Online platform data; causal attribution and external validity remain limited." }, { "id": "SRC-087", "role": "Counterevidence", "title": "Generative AI and Jobs: A 2025 Update", "authorOrOrganization": "International Labour Organization and NASK", "publicationDate": "2025-05-20", "url": "https://www.ilo.org/publications/generative-ai-and-jobs-2025-update", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Exposure mapping concludes job transformation is more likely than full replacement and does not measure realized job loss.", "limitations": "Exposure index is not a causal employment study." }, { "id": "SRC-R083", "role": "Counterevidence", "title": "The (Short-Term) Effects of Large Language Models on Unemployment and Earnings", "authorOrOrganization": "Chen, Kane, Kozlowski, Kunievsky, Evans (Univ. of Chicago Knowledge Lab)", "publicationDate": "2025-09-22", "url": "https://arxiv.org/abs/2509.15510", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Synthetic difference-in-differences finds earnings rose in highly LLM-exposed occupations after ChatGPT with no change in unemployment.", "limitations": "Preprint; occupation-level, not freelancers. Freelance-specific decline is well supported (Demirci et al., Management Science 2025), so this challenges only the earnings dimension and generality." } ] }, { "id": "I-004", "area": "Work & income", "direction": "Harm", "claim": "Evidence suggests early-2025 AI coding tools slowed experienced open-source developers working in their own mature repositories by about 19%, even though the developers believed the tools had sped them up.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Multi-country", "whoGains": "Analysis: none identified in this study; organizations that calibrate AI adoption to measured rather than perceived speedups", "whoBearsTheCost": "Experienced developers on familiar, mature codebases using early-2025 tools", "resolutionCriteria": "True if a follow-up independent RCT by 2028-12-31 again finds a slowdown of 10% or more; False if it finds a speedup of 10% or more; Partially true if the effect is within plus or minus 10%.", "ledgerEntry": "By 2028-12-31, independent RCTs will still show no net speedup of 10% or more for experienced developers on familiar mature codebases.", "sources": [ { "id": "SRC-R006", "role": "Supporting source", "title": "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity", "authorOrOrganization": "METR (Becker, Rush, Barnes, Rein)", "publicationDate": "2025-07-10", "url": "https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "RCT, 16 experienced developers, 246 tasks: AI-allowed tasks took 19% longer; developers expected a 24% speedup and afterward believed 20%.", "limitations": "Small sample; mostly Cursor with early-2025 models; not peer-reviewed." }, { "id": "SRC-R007", "role": "Supporting source", "title": "AI tools slow down experienced developers by 19%", "authorOrOrganization": "Diginomica", "publicationDate": "2025-07-15", "url": "https://diginomica.com/node/29061", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Trial ran February to June 2025; participants averaged five years and 1,500 commits on their projects.", "limitations": "Secondary summary." }, { "id": "SRC-011", "role": "Supporting source", "title": "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity", "authorOrOrganization": "METR", "publicationDate": "2025-07-10", "url": "https://arxiv.org/abs/2507.09089", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "In a randomized study of 16 experienced developers completing 246 tasks, AI access increased completion time by 19%.", "limitations": "Small sample, familiar repositories, and early-2025 tools." }, { "id": "SRC-088", "role": "Counterevidence", "title": "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity", "authorOrOrganization": "METR", "publicationDate": "2025-07-10", "url": "https://arxiv.org/abs/2507.09089", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "The randomized result covers 16 experienced maintainers, 246 tasks and early-2025 tools; it does not establish the effect for developers generally.", "limitations": "Small sample and a deliberately narrow population." }, { "id": "SRC-R085", "role": "Counterevidence", "title": "The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers (Management Science)", "authorOrOrganization": "Cui, Demirer, Jaffe, Musolff, Peng, Salz", "publicationDate": "2025-02-01", "url": "https://economics.mit.edu/sites/default/files/inline-files/draft_copilot_experiments.pdf", "tier": "Tier 1", "interestedSource": "Yes", "relevantFinding": "Randomized Copilot access raised completed tasks 26% across three firms; gains concentrated in junior developers, small for senior developers.", "limitations": "Interested (R7.4); different population (mixed seniority, own-company code) from METR." } ] }, { "id": "I-005", "area": "Work & income", "direction": "Gain", "claim": "Evidence suggests US workers who use generative AI report saving about 5.4% of their work hours, and self-reported time savings across all employed adults rose from 1.6% to 2.2% of work hours between Q3 2024 and Q2 2026.", "status": "Partial", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O1", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Frequent users, especially in computer and mathematics occupations and information services", "whoBearsTheCost": "Analysis: workers in firms that convert saved hours into reduced staffing (not sourced)", "resolutionCriteria": "True if the RPS time-savings series is at or above 3.0% of all work hours in its last 2028 release; False if at or below 1.6%; Partially true if between.", "ledgerEntry": "By 2028-12-31, self-reported generative AI time savings will reach at least 3.0% of all US work hours at Broad deployment reach (O1).", "sources": [ { "id": "SRC-R008", "role": "Supporting source", "title": "The Impact of Generative AI on Work Productivity", "authorOrOrganization": "Bick, Blandin, Deming (Federal Reserve Bank of St. Louis)", "publicationDate": null, "url": "https://www.stlouisfed.org/on-the-economy/2025/feb/impact-generative-ai-work-productivity?", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Users saved 5.4% of work hours on average (Nov 2024); 1.4% of all work hours including non-users.", "limitations": "Self-reported time savings. Published Feb 2025 (day not captured)." }, { "id": "SRC-R009", "role": "Supporting source", "title": "Does generative AI save time at work? (FRED Blog)", "authorOrOrganization": "Federal Reserve Bank of St. Louis", "publicationDate": null, "url": "https://fredblog.stlouisfed.org/2026/08/does-generative-ai-save-time-at-work/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Weekly work use rose from 28.2% (Q3 2024) to 39.2% (Q2 2026); hours saved rose from 1.6% to 2.2% of work hours.", "limitations": "Self-reported. Published Aug 2026 (day not captured)." }, { "id": "SRC-R048", "role": "Supporting source", "title": "AI's productivity is finally hitting the real economy", "authorOrOrganization": "SmartBrief (summarizing St. Louis Fed research)", "publicationDate": null, "url": "https://www.smartbrief.com/original/ais-productivity-is-finally-hitting-the-real-economy", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "St. Louis Fed estimates time savings of 1.6% of all work hours imply about a 1.3% labor productivity boost since ChatGPT.", "limitations": "Secondary; primary Fed analysis not fetched. Date not captured." }, { "id": "SRC-089", "role": "Counterevidence", "title": "Large Language Models, Small Labor Market Effects", "authorOrOrganization": "National Bureau of Economic Research", "publicationDate": "2025-04-01", "url": "https://www.nber.org/papers/w33777", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Linked employer-employee data found rapid adoption but no detectable effects on earnings or recorded hours through the study period.", "limitations": "National setting and early observation window may miss later or platform-specific effects." }, { "id": "SRC-113", "role": "Counterevidence", "title": "Large Language Models, Small Labor Market Effects", "authorOrOrganization": "National Bureau of Economic Research", "publicationDate": "2025-04-01", "url": "https://www.nber.org/papers/w33777", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Linked employer-employee data found rapid adoption but no detectable effects on earnings or recorded hours through the study period.", "limitations": "National setting and early observation window may miss later or platform-specific effects." } ] }, { "id": "I-006", "area": "Work & income", "direction": "Harm", "claim": "Unemployment among recent US college graduates could rise further by 2028 if AI substitutes for entry-level work; AI's contribution to the current elevated rate is not established.", "status": "Watch", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O0", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Not yet identifiable", "whoBearsTheCost": "Recent college graduates aged 22 to 27, especially in AI-exposed fields", "resolutionCriteria": "True if the rate is at or above 6.5% by the last 2028 release and at least one independent study attributes part of the rise to AI; False if the rate is at or below 4.5% or independent studies find no AI contribution; Partially true if the rate rises but AI attribution is not established.", "ledgerEntry": "By 2028-12-31, recent-graduate unemployment will be at or above 6.5% with at least one independent study attributing part of the rise to AI.", "sources": [ { "id": "SRC-R010", "role": "Supporting source", "title": "The Labor Market for Recent College Graduates", "authorOrOrganization": "Federal Reserve Bank of New York", "publicationDate": null, "url": "https://www.newyorkfed.org/research/college-labor-market?os=wi", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Recent-graduate unemployment about 5.6% through Q2 2026; underemployment about 42%.", "limitations": "Not AI-specific. Quarterly updates." }, { "id": "SRC-R011", "role": "Supporting source", "title": "Media Advisory: Research on the Role of Remote Work in Youth Unemployment", "authorOrOrganization": "Federal Reserve Bank of New York", "publicationDate": "2026-05-28", "url": "https://www.newyorkfed.org/newsevents/mediaadvisory/2026/0528-2026", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "March 2026 recent-grad unemployment 5.6% vs. 3.6% in March 2019; NY Fed examining remote work and generative AI as causes.", "limitations": "Advisory only; the underlying analysis was not fetched." }, { "id": "SRC-090", "role": "Counterevidence", "title": "Tracking the Impact of AI on the Labor Market", "authorOrOrganization": "Yale Budget Lab", "publicationDate": "2026-04-16", "url": "https://budgetlab.yale.edu/research/tracking-impact-ai-labor-market", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Updated analysis finds occupational mix, employment and unemployment have not yet changed in a way clearly connected to AI exposure.", "limitations": "Aggregate analysis may miss localized and age-specific effects." } ] }, { "id": "I-007", "area": "Work & income", "direction": "Gain", "claim": "Generative AI reduces completion time for well-defined professional writing tasks within the model's tested capability without lowering quality.", "status": "Confirmed", "currentReach": "Pilot", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Knowledge workers and organizations", "whoBearsTheCost": "Workers whose routine writing is commoditized", "resolutionCriteria": "By 2028, at least two independent field studies reproduce >=25% time savings without quality loss.", "ledgerEntry": "Professional writing AI cuts quality-adjusted task time >=25% by EOY 2028.", "sources": [ { "id": "SRC-003", "role": "Supporting source", "title": "Experimental evidence on the productivity effects of generative artificial intelligence", "authorOrOrganization": "Science / Noy and Zhang", "publicationDate": "2023-07-13", "url": "https://www.science.org/doi/10.1126/science.adh2586", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Randomized access reduced professional writing time by about 40% and increased output quality by about 18%.", "limitations": "Short, self-contained writing tasks; not long-run job performance." }, { "id": "SRC-R081", "role": "Counterevidence", "title": "Navigating the Jagged Technological Frontier (HBS WP 24-013, forthcoming Organization Science)", "authorOrOrganization": "Dell'Acqua, McFowland, Mollick, Lifshitz-Assaf, Kellogg, Rajendran, Krayer, Candelon, Lakhani", "publicationDate": "2023-09-15", "url": "https://papers.ssrn.com/abstract=4573321", "tier": "Tier 2", "interestedSource": "Yes", "relevantFinding": "Inside the AI frontier consultants were faster with higher quality, but on a task outside the frontier AI users were 19 percentage points less likely to reach the correct answer.", "limitations": "Preregistered; consultants not typical workers." } ] }, { "id": "I-008", "area": "Work & income", "direction": "Gain", "claim": "Evidence suggests job postings requiring AI skills offer a wage premium of about 5% within the same firm and job title.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O2", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Multi-country", "whoGains": "Workers able to acquire and signal scarce AI skills", "whoBearsTheCost": "Workers lacking training access or credential signals", "resolutionCriteria": "By 2028, matched worker or vacancy data establishes whether a >=5% premium persists after controls.", "ledgerEntry": "Verified AI skills carry >=5% within-role wage premium by EOY 2028.", "sources": [ { "id": "SRC-005", "role": "Supporting source", "title": "Artificial intelligence, job quality and inclusiveness", "authorOrOrganization": "OECD", "publicationDate": "2023-07-11", "url": "https://www.oecd.org/en/publications/oecd-employment-outlook-2023_08785bba-en/full-report/artificial-intelligence-job-quality-and-inclusiveness_a713d0ad.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Job postings requiring AI skills showed wage premiums, including about 5% within firm and job title.", "limitations": "Posting-based and predates widespread generative AI adoption." }, { "id": "SRC-R082", "role": "Counterevidence", "title": "Large Language Models, Small Labor Market Effects (NBER WP 33777)", "authorOrOrganization": "Humlum, Vestergaard", "publicationDate": "2025-05-09", "url": "https://ideas.repec.org/p/nbr/nberwo/33777.html", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Linking chatbot adoption surveys (25,000 workers, 7,000 workplaces, 11 exposed occupations) to administrative records, difference-in-differences finds precise null effects on earnings and hours, ruling out effects above 1 to 2%; time savings about 3%.", "limitations": "Denmark only; 2023 to 2024 tools; working paper." } ] }, { "id": "I-009", "area": "Work & income", "direction": "Harm", "claim": "Employment in highly AI-exposed occupations may fall by at least 2% relative to less-exposed occupations by 2028.", "status": "Watch", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O0", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Firms and customers if output costs fall", "whoBearsTheCost": "Displaced workers and affected communities", "resolutionCriteria": "By 2028, official labor data and a predeclared exposure design confirm or reject a >=2% relative decline.", "ledgerEntry": "Highly AI-exposed occupations lose >=2% employment relative to controls by EOY 2028.", "sources": [ { "id": "SRC-009", "role": "Supporting source", "title": "Generative AI and Jobs: A 2025 Update", "authorOrOrganization": "International Labour Organization and NASK", "publicationDate": "2025-05-20", "url": "https://www.ilo.org/publications/generative-ai-and-jobs-2025-update", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "One in four jobs has some GenAI exposure, but transformation is more likely than full replacement.", "limitations": "Exposure index, not an observed employment outcome." }, { "id": "SRC-R084", "role": "Counterevidence", "title": "AI Is Probably Not (Yet) the Reason for Labor Market Weakening", "authorOrOrganization": "The Budget Lab at Yale", "publicationDate": "2026-05-06", "url": "https://budgetlab.yale.edu/node/1495/pdf", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Econometric comparison of AI-exposed and unexposed occupations through early 2026 shows no clear AI labor-market effect; occupational mix change within historical range.", "limitations": "Aggregate; cannot rule out effects on early-career subgroups (contrast CLAIM-002)." } ] }, { "id": "I-010", "area": "Health & medicine", "direction": "Gain", "claim": "In a randomized trial of 105,915 women in Sweden's population screening program, AI-supported mammography detected more cancers, raised sensitivity from 73.8% to 80.5% at equal specificity, cut screen-reading workload by 44%, and was non-inferior on interval cancers.", "status": "Confirmed", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Women in screening programs (more invasive cancers found, fewer aggressive interval cancers); radiologists (lower reading workload)", "whoBearsTheCost": "Analysis: women in programs adopting AI without equivalent validation; trial covered one AI system, one vendor, one screening round", "resolutionCriteria": "True if by 2028-12-31 a second trial outside Sweden confirms non-inferior or better interval-cancer rates; False if a trial finds significantly worse rates; Partially true if only detection-rate (not interval-cancer) replication exists.", "ledgerEntry": "By 2028-12-31, AI-supported mammography will show replicated non-inferior interval-cancer outcomes in at least one additional country at Pilot reach or above (O3).", "sources": [ { "id": "SRC-R012", "role": "Supporting source", "title": "MASAI trial: interval cancer, sensitivity and specificity (The Lancet, 2026)", "authorOrOrganization": "Gommers, Lang et al. (Lund University)", "publicationDate": "2026-01-31", "url": "https://www.lunduniversity.lu.se/article/ai-support-breast-cancer-screening-fewer-missed-cancer-cases", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "105,915 women randomized; sensitivity 80.5% vs. 73.8% at 98.5% specificity; interval cancer 1.55 vs. 1.76 per 1,000 (non-inferior); earlier reports: 29% more cancers detected, 44% less reading workload.", "limitations": "Single program, one AI system, one vendor, one round; radiologists not masked. Link is the university's summary; DOI 10.1016/S0140-6736(25)02464-X." }, { "id": "SRC-R013", "role": "Supporting source", "title": "AI-supported mammography screening demonstrates non-inferior interval cancer rates", "authorOrOrganization": "Hospital Healthcare Europe", "publicationDate": null, "url": "https://hospitalhealthcare.com/clinical/oncology/ai-supported-mammography-screening-demonstrates-non-inferior-interval-cancer-rates/", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Trial conducted at routine population-based screening in southwest Sweden; authors note limited generalisability.", "limitations": "Secondary. Publication date not captured." }, { "id": "SRC-013", "role": "Supporting source", "title": "Artificial intelligence-supported screen reading versus standard double reading in the Mammography Screening with Artificial Intelligence trial", "authorOrOrganization": "Lancet Oncology / MASAI investigators", "publicationDate": "2023-08-01", "url": "https://pubmed.ncbi.nlm.nih.gov/37541274/", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Randomized trial cut screen-reading workload by 44.3% while maintaining a comparable cancer-detection rate at interim analysis.", "limitations": "One national screening context; long-term interval-cancer outcomes were not in the interim report." }, { "id": "SRC-091", "role": "Counterevidence", "title": "Artificial intelligence-supported screen reading versus standard double reading in mammography", "authorOrOrganization": "MASAI investigators / PubMed", "publicationDate": "2023-08-01", "url": "https://pubmed.ncbi.nlm.nih.gov/37541274/", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Interim results came from one national program, one workflow and one AI system; transfer to other populations and vendors remains unproven.", "limitations": "The trial itself supports efficacy in its stated population." }, { "id": "SRC-R086", "role": "Counterevidence", "title": "Use of artificial intelligence for image analysis in breast cancer screening programmes: systematic review of test accuracy (BMJ 2021;374:n1872)", "authorOrOrganization": "Freeman et al.", "publicationDate": "2021-09-01", "url": "https://www.bmj.com/content/bmj/374/bmj.n1872.full.pdf", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "In three large retrospective studies, 34 of 36 AI systems (94%) were less accurate than a single radiologist and all were less accurate than double reading.", "limitations": "Predates MASAI; superseded for this claim by the MASAI RCT. Verified 2026-09-23." } ] }, { "id": "I-011", "area": "Health & medicine", "direction": "Gain", "claim": "Evidence suggests ambient AI scribes modestly reduce physicians' note-writing time (about 9.5% versus usual care for one of two tools tested) with possible improvements in burnout measures.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Sub-national", "whoGains": "Outpatient physicians with high documentation burden", "whoBearsTheCost": "Analysis: patients if AI-drafted notes contain unreviewed errors; physicians using the tool that showed no significant time saving", "resolutionCriteria": "True if a non-deployer RCT by 2028-12-31 confirms time savings of 10% or more; False if it finds no saving; Partially true if savings are below 10%.", "ledgerEntry": "By 2028-12-31, an independent (non-deployer) randomized trial will confirm ambient AI scribes reduce physician documentation time by at least 10%.", "sources": [ { "id": "SRC-R014", "role": "Supporting source", "title": "Ambient AI Scribes in Clinical Practice: A Randomized Trial (NEJM AI, 2025)", "authorOrOrganization": "Lukac et al. (UCLA Health)", "publicationDate": "2025-11-26", "url": "https://uclahealth.org/news/release/ucla-study-finds-ai-scribes-may-reduce-documentation-time", "tier": "Tier 1", "interestedSource": "Yes", "relevantFinding": "238 physicians randomized to two AI scribes or usual care; one tool cut time-in-note 9.5% vs. control; the other not significant; possible burnout improvements.", "limitations": "Two-month trial, single center. Link is the deployer's news release." }, { "id": "SRC-R015", "role": "Supporting source", "title": "Studies find AI technology for clinical documentation aids efficiency and reduces burnout", "authorOrOrganization": "UW School of Medicine and Public Health / UW Health", "publicationDate": "2025-12-12", "url": "https://uwclinicaltrials.org/2025/12/12/studies-find-ai-technology-for-clinical-documentation-aids-efficiency-and-reduces-burnout/", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Randomized trial published in NEJM AI reports reduced documentation time and burnout.", "limitations": "Deployer's news summary; effect sizes not captured." }, { "id": "SRC-017", "role": "Supporting source", "title": "Evaluation of an Ambient Artificial Intelligence Documentation Platform for Clinicians", "authorOrOrganization": "JAMA Network Open", "publicationDate": "2025-05-01", "url": "https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2833433", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Ambient documentation use was associated with large reductions in reported mental demand, rushed pace, and effort, with improved documentation quality.", "limitations": "Nonrandomized implementation; self-selection and concurrent changes may contribute." }, { "id": "SRC-092", "role": "Counterevidence", "title": "Evaluation of an Ambient Artificial Intelligence Documentation Platform for Clinicians", "authorOrOrganization": "JAMA Network Open", "publicationDate": "2025-05-01", "url": "https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2833433", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Nonrandomized evidence reports burden reductions but cannot rule out selection, novelty or concurrent workflow changes.", "limitations": "Different product and design from the cited randomized trials." }, { "id": "SRC-R087", "role": "Counterevidence", "title": "UCLA study finds AI scribes may reduce documentation time (NEJM AI randomized trial, 2025)", "authorOrOrganization": "Lukac et al. (UCLA Health)", "publicationDate": "2025-11-26", "url": "https://uclahealth.org/news/release/ucla-study-finds-ai-scribes-may-reduce-documentation-time", "tier": "Tier 1", "interestedSource": "Yes", "relevantFinding": "238-physician RCT: one scribe cut note time 9.5% vs control; the other showed no significant difference; burnout effects suggestive only.", "limitations": "Two-month trial; single center." } ] }, { "id": "I-012", "area": "Health & medicine", "direction": "Harm", "claim": "Evidence suggests routine use of AI polyp detection may erode endoscopists' unassisted skill: adenoma detection in non-AI colonoscopies fell from 28.4% to 22.4% after AI was introduced at four Polish centres.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O2", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Analysis: patients during AI-assisted procedures (CADe raises detection per prior trials, not sourced this round)", "whoBearsTheCost": "Patients undergoing unassisted colonoscopy by AI-exposed endoscopists", "resolutionCriteria": "True if a prospective or controlled study by 2028-12-31 confirms an unassisted detection decline of 3 points or more; False if it finds no decline; Partially true if a smaller decline is found.", "ledgerEntry": "By 2028-12-31, a prospective study will confirm that sustained AI exposure lowers unassisted adenoma detection by at least 3 percentage points.", "sources": [ { "id": "SRC-R016", "role": "Supporting source", "title": "Endoscopist deskilling risk after exposure to AI in colonoscopy (Lancet Gastroenterol Hepatol 2025;10:896-903)", "authorOrOrganization": "Budzyn, Romanczyk et al.", "publicationDate": "2025-08-12", "url": "https://doi.org/10.1016/S2468-1253(25)00133-5", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Unassisted ADR fell from 28.4% to 22.4% after AI introduction; AI exposure independent predictor (OR 0.69).", "limitations": "Retrospective observational; 3-month windows; 19 endoscopists." }, { "id": "SRC-R017", "role": "Supporting source", "title": "Artificial Intelligence in Colonoscopy: Could It Be Making Us Worse? (Evidence-Based GI)", "authorOrOrganization": "American College of Gastroenterology", "publicationDate": "2025-09-17", "url": "https://gi.org/journals-publications/ebgi/zhou_sep2025/", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Four centres in the ACCEPT trial; summary notes guideline bodies rate evidence on CADe long-term outcomes as very low certainty.", "limitations": "Secondary commentary." }, { "id": "SRC-093", "role": "Counterevidence", "title": "Artificial intelligence-assisted colonoscopy: updated systematic review and meta-analysis", "authorOrOrganization": "Independent academic investigators", "publicationDate": "2023-01-01", "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12851087/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Randomized evidence generally finds higher adenoma detection with AI assistance, so one short before-after deskilling signal is not the full clinical record.", "limitations": "Assisted detection benefit does not directly test unassisted skill after exposure." } ] }, { "id": "I-013", "area": "Health & medicine", "direction": "Gain", "claim": "Evidence suggests point-of-care autonomous AI eye exams sharply increase diabetic eye exam completion among youth with diabetes (100% vs. 22% within six months in a randomized trial of 164 participants).", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Sub-national", "whoGains": "Youth with diabetes, including racially and ethnically diverse and underserved patients with low baseline screening rates", "whoBearsTheCost": "Analysis: patients with false-negative AI results; clinics bearing device costs (not sourced)", "resolutionCriteria": "True if an independent RCT by 2028-12-31 confirms a completion gain of 20 points or more; False if it finds no gain; Partially true if the gain is smaller.", "ledgerEntry": "By 2028-12-31, an independent randomized trial will confirm autonomous AI eye exams raise diabetic eye exam completion by at least 20 percentage points.", "sources": [ { "id": "SRC-R018", "role": "Supporting source", "title": "Autonomous AI increases screening and follow-up for diabetic retinopathy in youth: the ACCESS RCT (Nature Communications 15:421)", "authorOrOrganization": "Wolf, Channa, Liu ... Abramoff", "publicationDate": "2024-01-11", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10784572/", "tier": "Tier 1", "interestedSource": "Yes", "relevantFinding": "164 youth randomized; exam completion 100% vs. 22%; follow-through after abnormal result 64% vs. 22%.", "limitations": "Single center; small sample; vendor-affiliated author." }, { "id": "SRC-015", "role": "Supporting source", "title": "Autonomous artificial intelligence increases screening and follow-up for diabetic retinopathy in youth", "authorOrOrganization": "Nature Communications / ACCESS investigators", "publicationDate": "2024-01-18", "url": "https://www.nature.com/articles/s41467-023-44676-z", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Randomized point-of-care autonomous AI increased completed diabetic-eye screening and follow-up in a diverse youth population.", "limitations": "Single health-system setting and specific disease workflow." }, { "id": "SRC-094", "role": "Counterevidence", "title": "Autonomous artificial intelligence increases screening and follow-up for diabetic retinopathy in youth", "authorOrOrganization": "Nature Communications", "publicationDate": "2024-01-18", "url": "https://www.nature.com/articles/s41467-023-44676-z", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "The trial was single-center, small and included a device-affiliated author; completion does not establish improved vision outcomes.", "limitations": "Directly supports the process outcome in that setting." } ] }, { "id": "I-014", "area": "Health & medicine", "direction": "Harm", "claim": "Large Medicare Advantage insurers' post-acute prior-authorization denial rates rose sharply (UnitedHealthcare from 8.7% in 2019 to 22.7% in 2022) while they deployed automated and predictive review tools; AI's contribution to the increase could be material but is not established.", "status": "Watch", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O2", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Insurers (lower post-acute care spending); Analysis: none identified for patients", "whoBearsTheCost": "Medicare Advantage enrollees needing skilled nursing, inpatient rehabilitation or long-term acute care after hospitalization", "resolutionCriteria": "True if by 2028-12-31 an official source shows the largest insurers' post-acute denial rates remain 10 or more points above smaller insurers and at least one official source ties the gap partly to automated review; False if the gap falls within 5 points; Partially true otherwise.", "ledgerEntry": "By 2028-12-31, large Medicare Advantage insurers' post-acute denial rates will remain at least 10 points above smaller insurers', with automated review cited as a contributing factor by an official source.", "sources": [ { "id": "SRC-R019", "role": "Supporting source", "title": "Refusal of Recovery: How Medicare Advantage Insurers Have Denied Patients Access to Post-Acute Care", "authorOrOrganization": "US Senate Permanent Subcommittee on Investigations (Majority Staff)", "publicationDate": "2024-10-17", "url": "https://www.hsgac.senate.gov/wp-content/uploads/2024.10.17-PSI-Majority-Staff-Report-on-Medicare-Advantage.pdf", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Post-acute denial rates rose sharply 2019 to 2022 at the three largest insurers; section on algorithms, AI and automation.", "limitations": "Majority-staff report; insurers dispute characterization." }, { "id": "SRC-R020", "role": "Supporting source", "title": "Senate Report Shows How Medicare Advantage Giants Lie to Seniors and Deny Health Care", "authorOrOrganization": "Center for Economic and Policy Research", "publicationDate": null, "url": "https://cepr.net/senate-report-shows-how-medicare-advantage-giants-lie-to-seniors-and-deny-health-care/", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "UnitedHealthcare, Humana and CVS covered 59% of MA enrollees in 2024; UHC post-acute denial rate 8.7% (2019) to 22.7% (2022); internal notes touted machine-assisted prior authorization raising adverse determinations.", "limitations": "Advocacy framing; used for figures only. Publication date not captured." }, { "id": "SRC-R067", "role": "Supporting source", "title": "Letter to UnitedHealthcare CEO (July 14, 2026)", "authorOrOrganization": "US Senate Permanent Subcommittee on Investigations", "publicationDate": "2026-07-14", "url": "https://www.hsgac.senate.gov/wp-content/uploads/2026-07-14-letter-to-UHC.pdf", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Cites HHS OIG data: UnitedHealthcare denied 66% of inpatient rehabilitation admission requests, well above smaller insurers.", "limitations": "Letter summarizing OIG reports; OIG reports not fetched." }, { "id": "SRC-095", "role": "Counterevidence", "title": "Medicare Advantage insurers made nearly 53 million prior authorization determinations in 2024", "authorOrOrganization": "KFF", "publicationDate": "2026-02-01", "url": "https://www.kff.org/medicare/medicare-advantage-insurers-made-nearly-53-million-prior-authorization-determinations-in-2024/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Current insurer denial rates vary substantially and do not move uniformly with automation, weakening causal attribution from coincident deployment.", "limitations": "Public data do not identify each decision technology." } ] }, { "id": "I-015", "area": "Health & medicine", "direction": "Harm", "claim": "Use of general-purpose chatbots for emotional support by teens could expose vulnerable young users to harmful responses; population-level mental-health effects are not yet measured.", "status": "Watch", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O0", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Not yet identifiable", "whoBearsTheCost": "Not yet identifiable; Analysis: teens in distress relying on chatbots instead of human support", "resolutionCriteria": "True if by 2028-12-31 a controlled or population study finds harm attributable to teen chatbot use for emotional support; False if such a study finds no harm or benefit; Partially true if findings are mixed across subgroups.", "ledgerEntry": "By 2028-12-31, a controlled or population study will find measurable harm attributable to teens' use of chatbots for emotional support.", "sources": [ { "id": "SRC-R021", "role": "Supporting source", "title": "Over Half of US Teens Use AI for Schoolwork, Pew Data Shows", "authorOrOrganization": "eWeek (reporting Pew Research Center)", "publicationDate": "2026-02-25", "url": "https://www.eweek.com/news/teens-ai-chatbot-use-pew-survey/", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "About two-thirds of teens have used chatbots; 12% used them for emotional support.", "limitations": "Secondary report of Pew; primary report not fetched." }, { "id": "SRC-096", "role": "Counterevidence", "title": "Systematic review of conversational agents for mental health", "authorOrOrganization": "npj Digital Medicine", "publicationDate": "2023-01-01", "url": "https://www.nature.com/articles/s41746-023-00883-5", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Some controlled studies report short-term symptom or engagement benefits, so harmful effects cannot be inferred from exposure alone.", "limitations": "Heterogeneous, mostly short studies; teen-specific evidence is limited." } ] }, { "id": "I-016", "area": "Health & medicine", "direction": "Harm", "claim": "Evidence suggests a widely deployed proprietary sepsis prediction model, in the version evaluated in 2021, missed about two-thirds of sepsis cases at one external health system.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Health systems that learn to demand local validation", "whoBearsTheCost": "Patients and clinicians relying on poorly calibrated alerts", "resolutionCriteria": "By 2028, version-specific multicenter external validations establish whether broad deployments clear safety thresholds.", "ledgerEntry": "Broad sepsis AI deployments achieve >=80% sensitivity with acceptable alert burden by EOY 2028.", "sources": [ { "id": "SRC-019", "role": "Supporting source", "title": "External Validation of a Widely Implemented Proprietary Sepsis Prediction Model", "authorOrOrganization": "JAMA Internal Medicine", "publicationDate": "2021-06-21", "url": "https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2781307", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "At one external health system, the model had AUC 0.63 and missed 67% of sepsis cases at the evaluated threshold. The article also documents that the model was already implemented at hundreds of US hospitals.", "limitations": "One implementation and earlier model version; vendors and workflows can improve." } ] }, { "id": "I-017", "area": "Health & medicine", "direction": "Harm", "claim": "AI-related software or algorithm failures in medical devices may cause verified serious patient harm by 2028.", "status": "Watch", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O1", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Patients if stronger surveillance prevents recurrence", "whoBearsTheCost": "Patients exposed to unsafe device behavior", "resolutionCriteria": "By 2028, regulator findings determine whether >=10 serious injuries have primary algorithmic causation.", "ledgerEntry": "Medical AI causes >=10 regulator-adjudicated serious injuries across two device classes by EOY 2028.", "sources": [ { "id": "SRC-021", "role": "Supporting source", "title": "As AI enters the operating room, reports arise of botched surgeries and misidentified body parts", "authorOrOrganization": "Reuters", "publicationDate": "2026-02-09", "url": "https://www.reuters.com/investigations/ai-enters-operating-room-reports-arise-botched-surgeries-misidentified-body-2026-02-09/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Investigation identified 1,401 adverse-event reports for selected devices, with 115 mentioning software or algorithm issues.", "limitations": "Reports do not prove AI caused each injury and are subject to reporting bias." } ] }, { "id": "I-018", "area": "Education & skills", "direction": "Gain", "claim": "Evidence suggests a pedagogically designed AI tutor produced more than twice the learning gains of an in-class active-learning session, in less time, in a randomized crossover trial with 194 Harvard physics students.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Sub-national", "whoGains": "Undergraduates in the studied course; Analysis: students who need self-paced instruction", "whoBearsTheCost": "Analysis: none identified in the study; generalization to less-selective institutions untested", "resolutionCriteria": "True if an independent RCT at another institution by 2028-12-31 finds significantly higher learning gains; False if it finds none; Partially true if gains are positive but not significant.", "ledgerEntry": "By 2028-12-31, an independent randomized trial at another institution will replicate significantly higher learning gains from a pedagogically designed AI tutor.", "sources": [ { "id": "SRC-R022", "role": "Supporting source", "title": "AI tutoring outperforms in-class active learning: an RCT (Scientific Reports 15:17458)", "authorOrOrganization": "Kestin, Miller, Klales, Milbourne, Ponti", "publicationDate": "2025-06-03", "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12179260/", "tier": "Tier 1", "interestedSource": "Yes", "relevantFinding": "194 students; AI tutor produced more than twice the learning gains in less time than in-class active learning.", "limitations": "Two lessons; highly selective institution." }, { "id": "SRC-097", "role": "Counterevidence", "title": "OECD Digital Education Outlook 2026", "authorOrOrganization": "OECD", "publicationDate": "2026-01-20", "url": "https://www.oecd.org/en/publications/oecd-digital-education-outlook-2026_062a7394-en.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Learning effects depend on pedagogy: guided use can help while task outsourcing can improve performance without learning.", "limitations": "Cross-country synthesis; evidence remains young." }, { "id": "R66-E21-1", "role": "Counterevidence", "title": "University AI tutoring trial", "authorOrOrganization": "Liu et al., University of Maryland", "publicationDate": "2026-09-30", "url": "https://edworkingpapers.com/sites/default/files/ai26-1598.pdf", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Within-course access estimates: grades -0.37 SD; LMS participation -0.90 SD. Full-sample grade effect not significant. Three nonactivating instructors excluded after assignment.", "limitations": "Low uptake; course grades and platform logs do not establish durable skill loss. Claimed ITT is limited by post-assignment exclusions. Plan registered after collection, before analysis." } ], "lastReviewed": "2026-10-06" }, { "id": "I-019", "area": "Education & skills", "direction": "Harm", "claim": "In a field experiment with nearly 1,000 Turkish high school math students, unrestricted GPT-4 access during practice lowered later unassisted exam scores by 17% compared with no access, while a tutor with learning safeguards largely avoided the loss.", "status": "Confirmed", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Analysis: students given guardrailed tutors (no exam loss, large practice gains)", "whoBearsTheCost": "Secondary students using unguarded chatbots as a crutch during practice", "resolutionCriteria": "True if by 2028-12-31 an independent study in another setting confirms a loss of 10% or more; False if it finds no loss; Partially true if a smaller loss is found.", "ledgerEntry": "By 2028-12-31, an independent study will confirm that unguarded chatbot practice reduces unassisted exam performance by at least 10%.", "sources": [ { "id": "SRC-R023", "role": "Supporting source", "title": "Generative AI without guardrails can harm learning: Evidence from high school mathematics (PNAS 122(26))", "authorOrOrganization": "Bastani, Bastani, Sungu, Ge, Kabakci, Mariman", "publicationDate": "2025-07-01", "url": "https://ideas.repec.org/a/nas/journl/v122y2025pe2422633122.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "About 1,000 students; GPT Base improved practice 48% but cut exam scores 17% vs. control; GPT Tutor improved practice 127% with no exam loss.", "limitations": "One school; GPT-4 era tools." }, { "id": "SRC-R024", "role": "Supporting source", "title": "Without Guardrails, Generative AI Can Harm Education", "authorOrOrganization": "Knowledge at Wharton", "publicationDate": null, "url": "https://knowledge.wharton.upenn.edu/article/without-guardrails-generative-ai-can-harm-education", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Experiment with nearly 1,000 high school math students in Turkey across three arms.", "limitations": "Institutional summary. Publication date not captured." }, { "id": "SRC-027", "role": "Supporting source", "title": "Generative AI without guardrails can harm learning", "authorOrOrganization": "Proceedings of the National Academy of Sciences", "publicationDate": "2025-06-01", "url": "https://www.pnas.org/doi/10.1073/pnas.2422633122", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Field experiment found unrestricted GPT improved practice performance but reduced later unaided performance; guarded tutoring avoided that harm.", "limitations": "One math course and implementation; published correction should be read with the article." }, { "id": "SRC-029", "role": "Supporting source", "title": "Generative AI without guardrails can harm learning", "authorOrOrganization": "Proceedings of the National Academy of Sciences", "publicationDate": "2025-06-01", "url": "https://www.pnas.org/doi/10.1073/pnas.2422633122", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Field experiment found unrestricted GPT improved practice performance but reduced later unaided performance; guarded tutoring avoided that harm.", "limitations": "One math course and implementation; published correction should be read with the article." }, { "id": "SRC-098", "role": "Counterevidence", "title": "Generative AI without guardrails can harm learning", "authorOrOrganization": "Proceedings of the National Academy of Sciences", "publicationDate": "2025-06-01", "url": "https://www.pnas.org/doi/10.1073/pnas.2422633122", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "The same experiment found a guarded tutor largely avoided the unaided-performance harm, limiting generalization from unrestricted use.", "limitations": "One school, subject and model generation." }, { "id": "SRC-R089", "role": "Counterevidence", "title": "Generative AI without guardrails can harm learning (PNAS 122(26))", "authorOrOrganization": "Bastani et al.", "publicationDate": "2025-07-01", "url": "https://ideas.repec.org/a/nas/journl/v122y2025pe2422633122.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "The guardrailed GPT Tutor raised practice scores 127% but produced no significant improvement on the later unassisted exam.", "limitations": "Same study as the supporting source." } ] }, { "id": "I-020", "area": "Education & skills", "direction": "Gain", "claim": "Evidence suggests a six-week, teacher-supervised GPT-4 after-school program raised Nigerian senior secondary students' test scores by 0.31 standard deviations (0.23 in English), with gains also on end-of-year exams.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Participating students, with larger gains for female and higher-performing students", "whoBearsTheCost": "Lower-performing and lower-socioeconomic students gained less, widening gaps (as summarized in SRC-R027)", "resolutionCriteria": "True if an independent RCT by 2028-12-31 finds gains of 0.2 SD or more; False if it finds none; Partially true if gains are smaller.", "ledgerEntry": "By 2028-12-31, an independent randomized trial in another low- or middle-income setting will replicate learning gains of at least 0.2 SD from supervised LLM tutoring.", "sources": [ { "id": "SRC-R025", "role": "Supporting source", "title": "From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria", "authorOrOrganization": "De Simone, Tiberti, Barron Rodriguez, Manolio, Mosuro, Dikoru (World Bank)", "publicationDate": "2025-05-20", "url": "https://hdl.handle.net/10986/43212", "tier": "Tier 2", "interestedSource": "Yes", "relevantFinding": "RCT: 0.31 SD overall gain; 0.23 SD in English over six weeks.", "limitations": "Working paper; short program." }, { "id": "SRC-R026", "role": "Supporting source", "title": "How AI tutors improved learning in Nigeria", "authorOrOrganization": "VoxDev (study authors)", "publicationDate": "2026-03-17", "url": "https://voxdev.org/node/300094", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Nine public secondary schools in Edo State; students also did better on end-of-year exams.", "limitations": "Author summary." }, { "id": "SRC-R027", "role": "Supporting source", "title": "World Bank Group: From Chalkboard to Chatbots (summary)", "authorOrOrganization": "ibl.ai", "publicationDate": "2025-06-11", "url": "https://ibl.ai/blog/world-bank-group-from-chalkboard-to-chatbots-evaluating-the-impact-of-generative-ai-on-learning-outcomes-in-nigeria", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Reports stronger gains for higher-performing, female and higher-socioeconomic students.", "limitations": "Vendor summary; verify against working paper." }, { "id": "SRC-099", "role": "Counterevidence", "title": "OECD Digital Education Outlook 2026", "authorOrOrganization": "OECD", "publicationDate": "2026-01-20", "url": "https://www.oecd.org/en/publications/oecd-digital-education-outlook-2026_062a7394-en.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Learning effects depend on pedagogy: guided use can help while task outsourcing can improve performance without learning.", "limitations": "Cross-country synthesis; evidence remains young." } ] }, { "id": "I-021", "area": "Education & skills", "direction": "Gain", "claim": "Evidence suggests real-time AI guidance for human tutors raised K-12 students' math topic mastery by 4 percentage points (9 points for students of lower-rated tutors) at about $20 per tutor per year, in a preregistered randomized trial in Title I schools.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Sub-national", "whoGains": "Students of less-experienced or lower-rated tutors in low-income schools", "whoBearsTheCost": "Analysis: none identified; tutors noted some suggestions were not grade-appropriate", "resolutionCriteria": "True if an independent RCT by 2028-12-31 finds gains of 3 points or more; False if none; Partially true if smaller.", "ledgerEntry": "By 2028-12-31, an independent randomized trial will replicate mastery gains of at least 3 percentage points from AI guidance for human tutors.", "sources": [ { "id": "SRC-R028", "role": "Supporting source", "title": "Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise", "authorOrOrganization": "Wang, Ribeiro, Robinson, Loeb, Demszky", "publicationDate": "2024-10-03", "url": "https://arxiv.org/abs/2410.03017", "tier": "Tier 2", "interestedSource": "Yes", "relevantFinding": "Preregistered RCT; students 4 points more likely to master topics; 9 points for lower-rated tutors; about $20 per tutor per year.", "limitations": "Preprint; sample figures revised in later version." }, { "id": "SRC-R029", "role": "Supporting source", "title": "Tutor CoPilot study page", "authorOrOrganization": "National Student Support Accelerator (Stanford)", "publicationDate": null, "url": "https://nssa.stanford.edu/studies/tutor-copilot-human-ai-approach-scaling-real-time-expertise", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Revised version: more than 700 tutors and 1,000 students.", "limitations": "Publication date not captured." }, { "id": "SRC-100", "role": "Counterevidence", "title": "OECD Digital Education Outlook 2026", "authorOrOrganization": "OECD", "publicationDate": "2026-01-20", "url": "https://www.oecd.org/en/publications/oecd-digital-education-outlook-2026_062a7394-en.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Learning effects depend on pedagogy: guided use can help while task outsourcing can improve performance without learning.", "limitations": "Cross-country synthesis; evidence remains young." } ] }, { "id": "I-022", "area": "Education & skills", "direction": "Harm", "claim": "Evidence suggests AI-writing detectors tested in 2023 misclassified essays by non-native English writers as AI-generated at high rates (a 61.3% average false-positive rate across seven detectors on 91 TOEFL essays), exposing those students to false accusations where such detectors were used.", "status": "Partial", "currentReach": "Lab", "projectedReachByEOY2028": "Lab", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Multi-country", "whoGains": "Analysis: none identified", "whoBearsTheCost": "Non-native English-writing students in institutions using AI detectors", "resolutionCriteria": "True if an independent audit by 2028-12-31 finds false-positive rates above 20% on current detectors; False if below 5%; Partially true if between.", "ledgerEntry": "By 2028-12-31, independent audits will still find false-positive rates above 20% for non-native English writing on widely used AI-writing detectors.", "sources": [ { "id": "SRC-R030", "role": "Supporting source", "title": "GPT detectors are biased against non-native English writers (Patterns, 2023)", "authorOrOrganization": "Liang, Yuksekgonul, Mao, Wu, Zou", "publicationDate": "2023-07-10", "url": "https://www.cell.com/patterns/fulltext/S2666-3899(23)00130-7", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Seven widely used detectors misclassified 91 TOEFL essays by non-native writers as AI-generated at an average false-positive rate of 61.3%, while US eighth-grade essays were classified accurately.", "limitations": "2023 detectors; peer-reviewed opinion article in Patterns. Verified 2026-09-23." }, { "id": "SRC-101", "role": "Counterevidence", "title": "Guidance for generative AI in education and research", "authorOrOrganization": "UNESCO", "publicationDate": "2023-09-01", "url": "https://unesdoc.unesco.org/ark:/48223/pf0000386693", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "UNESCO cautions that AI-detection tools are unreliable enough that disciplinary decisions require human evidence and due process.", "limitations": "Policy guidance rather than a current detector benchmark." } ] }, { "id": "I-023", "area": "Education & skills", "direction": "Harm", "claim": "Widespread teen use of chatbots for schoolwork could weaken independent skill-building where tools are used without learning safeguards; population-level learning effects are not yet measured.", "status": "Watch", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O0", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Not yet identifiable", "whoBearsTheCost": "Not yet identifiable; Analysis: students using chatbots to complete rather than learn", "resolutionCriteria": "True if by 2028-12-31 a population or multi-site controlled study links chatbot use to lower unassisted achievement; False if it finds no decline or gains; Partially true if effects depend on guardrails.", "ledgerEntry": "By 2028-12-31, a population or multi-site controlled study will link unguarded student chatbot use to lower unassisted achievement.", "sources": [ { "id": "SRC-R031", "role": "Supporting source", "title": "Over Half of US Teens Use AI for Schoolwork, Pew Data Shows", "authorOrOrganization": "eWeek (reporting Pew Research Center)", "publicationDate": "2026-02-25", "url": "https://www.eweek.com/news/teens-ai-chatbot-use-pew-survey/", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "More than half of teens report using chatbots for help with assignments or research.", "limitations": "Secondary report of Pew." }, { "id": "SRC-R032", "role": "Supporting source", "title": "About a quarter of U.S. teens have used ChatGPT for schoolwork", "authorOrOrganization": "Pew Research Center", "publicationDate": "2025-01-15", "url": "https://www.pewresearch.org/short-reads/2025/01/15/about-a-quarter-of-us-teens-have-used-chatgpt-for-schoolwork-double-the-share-in-2023/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "26% of teens used ChatGPT for schoolwork in fall 2024, up from 13% in 2023.", "limitations": "ChatGPT only; survey Sep to Oct 2024." }, { "id": "SRC-102", "role": "Counterevidence", "title": "AI accessibility performance gaps for disabled users: a review", "authorOrOrganization": "Independent review", "publicationDate": "2026-03-01", "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12980396/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Recurring performance gaps mean general tools cannot be assumed to produce learning or access gains without targeted data.", "limitations": "Heterogeneous systems and outcomes." }, { "id": "R66-E21-1", "role": "Context / limits", "title": "University AI tutoring trial", "authorOrOrganization": "Liu et al., University of Maryland", "publicationDate": "2026-09-30", "url": "https://edworkingpapers.com/sites/default/files/ai26-1598.pdf", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Within-course access estimates: grades -0.37 SD; LMS participation -0.90 SD. Full-sample grade effect not significant. Three nonactivating instructors excluded after assignment.", "limitations": "Low uptake; course grades and platform logs do not establish durable skill loss. Claimed ITT is limited by post-assignment exclusions. Plan registered after collection, before analysis." } ], "lastReviewed": "2026-10-06" }, { "id": "I-024", "area": "Education & skills", "direction": "Gain", "claim": "Guided generative AI reduces teacher lesson-planning time by at least 20% without lowering resource quality.", "status": "Confirmed", "currentReach": "Pilot", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Teachers and students", "whoBearsTheCost": "Teachers if generated errors add review work", "resolutionCriteria": "By 2028, independent trials in >=3 subjects reproduce the threshold.", "ledgerEntry": "Guided teacher AI cuts planning time >=20% without lower quality by EOY 2028.", "sources": [ { "id": "SRC-025", "role": "Supporting source", "title": "Using generative AI for KS3 science lesson preparation", "authorOrOrganization": "Education Endowment Foundation", "publicationDate": "2025-02-12", "url": "https://educationendowmentfoundation.org.uk/projects-and-evaluation/projects/choices-in-edtech-using-generative-ai-chatgpt-for-ks3-science-lesson-preparation-2024-teacher-choices-trial", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Randomized trial with 259 teachers found roughly 31% less planning time, about 25 minutes weekly, without detected quality loss.", "limitations": "Short horizon, one subject and age band; resource quality measures are not student outcomes. Workload time was self-reported through a researcher-designed weekly diary rather than passively timed. The primary result summed weeks 6–10; up to two missing weekly entries were within-teacher mean-imputed, and cases with more than two missing entries were excluded." }, { "id": "R66-E20-1", "role": "Supporting source", "title": "Aila teacher preparation trial", "authorOrOrganization": "NFER", "publicationDate": "2026-10-06", "url": "https://www.nfer.ac.uk/publications/lesson-planning-using-ai-lesson-assistant-aila-a-teacher-choices-trial/", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "464 KS2 teachers at 108 schools. Preparation time was 24% lower with Aila access; assessed resource quality was no worse. Use covered only 7% of lessons and declined.", "limitations": "Workload is self-reported rather than directly timed performance. Resource quality is not student achievement; the subject breakdown has not been established for the three-subject replication test." } ], "lastReviewed": "2026-10-06" }, { "id": "I-025", "area": "Education & skills", "direction": "Gain", "claim": "Evidence suggests generative AI assistance narrowed an education-based task-performance gap in one controlled study, without evidence yet of durable skill gains.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Adults with less formal education", "whoBearsTheCost": "Users if short-term assistance is mistaken for skill acquisition", "resolutionCriteria": "By 2028, delayed-assessment trials determine whether assisted gap narrowing persists.", "ledgerEntry": "At least half of AI-assisted education-gap narrowing persists unaided after 30 days by EOY 2028.", "sources": [ { "id": "SRC-031", "role": "Supporting source", "title": "Does Generative AI Narrow Education-Based Productivity Gaps?", "authorOrOrganization": "National Bureau of Economic Research", "publicationDate": "2026-04-01", "url": "https://www.nber.org/papers/w34851", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Randomized evidence found larger task-performance gains for less-educated participants, narrowing an observed productivity gap.", "limitations": "Task performance is not durable skill acquisition or labor-market mobility." }, { "id": "SRC-R090", "role": "Counterevidence", "title": "The Uneven Impact of Generative AI on Entrepreneurial Performance (Management Science)", "authorOrOrganization": "Otis, Clarke, Delecourt, Holtz, Koning", "publicationDate": "2026-07-10", "url": "https://escholarship.org/content/qt0fb001r8/qt0fb001r8.pdf", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Randomized AI business mentor: no average effect; high performers gained about 15% while low performers lost about 8%, widening gaps.", "limitations": "Different setting (business outcomes)." } ] }, { "id": "I-026", "area": "Education & skills", "direction": "Harm", "claim": "Unequal access and use may produce at least a 10-point socioeconomic gap in regular educational AI use.", "status": "Watch", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O1", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Multi-country", "whoGains": "Students with devices, connectivity, and guidance", "whoBearsTheCost": "Low-income students and under-resourced schools", "resolutionCriteria": "By 2028, representative student surveys establish whether the regular-use gap reaches 10 points.", "ledgerEntry": "Regular educational AI use differs >=10 points by socioeconomic group at EOY 2028.", "sources": [ { "id": "SRC-033", "role": "Supporting source", "title": "AI use by individuals surges across the OECD as adoption by firms continues to expand", "authorOrOrganization": "OECD", "publicationDate": "2026-01-28", "url": "https://www.oecd.org/en/about/news/announcements/2026/01/ai-use-by-individuals-surges-across-the-oecd-as-adoption-by-firms-continues-to-expand.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "More than one-third of individuals used generative AI in 2025, with large gaps by age, education, income, and firm size.", "limitations": "Usage does not reveal purpose, quality, or educational outcome." }, { "id": "SRC-R091", "role": "Counterevidence", "title": "About a quarter of U.S. teens have used ChatGPT for schoolwork", "authorOrOrganization": "Pew Research Center", "publicationDate": "2025-01-15", "url": "https://www.pewresearch.org/short-reads/2025/01/15/about-a-quarter-of-us-teens-have-used-chatgpt-for-schoolwork-double-the-share-in-2023/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Black and Hispanic teens (31%) were more likely than White teens (22%) to use ChatGPT for schoolwork.", "limitations": "Race, not income; ChatGPT only." } ] }, { "id": "I-027", "area": "Education & skills", "direction": "Gain", "claim": "AI tools may deliver measurable learning or access gains for students with disabilities by 2028.", "status": "Watch", "currentReach": "Pilot", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O1", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Students with disabilities and their teachers", "whoBearsTheCost": "Students poorly represented in training and evaluation", "resolutionCriteria": "By 2028, two independent controlled studies meet a prespecified educational or access threshold.", "ledgerEntry": "Two independent studies show meaningful learning or access gains from AI for disabled students by EOY 2028.", "sources": [ { "id": "SRC-035", "role": "Supporting source", "title": "Artificial intelligence and digital accessibility: a systematic review", "authorOrOrganization": "PubMed Central review", "publicationDate": "2024-02-01", "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10905618/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "AI can improve accessibility but cost, training data, usability, infrastructure, and co-design determine who benefits.", "limitations": "Predominantly early-stage studies, with limited population deployment evidence." }, { "id": "SRC-R092", "role": "Counterevidence", "title": "Community-Supported Shared Infrastructure in Support of Speech Accessibility (JSLHR)", "authorOrOrganization": "Hasegawa-Johnson et al.", "publicationDate": null, "url": "https://catalogue.yorvoice.york.ac.uk/catalogue/RQ3L34RR", "tier": "Tier 2", "interestedSource": "Yes", "relevantFinding": "Baseline speech recognition had 36.3% word error rate for speakers with Parkinson's vs 3.4% for typical speakers.", "limitations": "Adults, not students; improvement possible with targeted data." } ] }, { "id": "I-028", "area": "Science & discovery", "direction": "Gain", "claim": "Evidence suggests machine-learning weather models now run operationally at the European Centre for Medium-Range Weather Forecasts and outperform its physics-based system on many measures, including tropical cyclone tracks by up to 20%, using about 1,000 times less energy per forecast.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Multi-country", "whoGains": "Forecast users in ECMWF member and cooperating states (weather services, emergency planners)", "whoBearsTheCost": "Users relying on high-resolution fields, where the ML ensemble (31 km) remains coarser than the physics system (9 km) (SRC-R033)", "resolutionCriteria": "True if an independent verification by 2028-12-31 finds ML models match or beat physics models on most headline scores; False if they underperform; Partially true if mixed.", "ledgerEntry": "By 2028-12-31, independent verification will confirm operational ML weather models match or beat physics-based models on most headline scores.", "sources": [ { "id": "SRC-R033", "role": "Supporting source", "title": "ECMWF's ensemble AI forecasts become operational", "authorOrOrganization": "ECMWF", "publicationDate": "2025-07-01", "url": "https://www.ecmwf.int/en/about/media-centre/news/2025/ecmwfs-ensemble-ai-forecasts-become-operational", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "AIFS ENS operational; outperforms physics models on many measures by up to 20%; about 1,000 times less energy; runs at 31 km vs. 9 km physics ensemble.", "limitations": "Self-reported; single-model operational since Feb 2025." }, { "id": "SRC-R034", "role": "Supporting source", "title": "AIFS Single v1.1 (EGUsphere preprint)", "authorOrOrganization": "ECMWF authors", "publicationDate": "2025-10-17", "url": "https://egusphere.copernicus.org/preprints/2025/egusphere-2025-4716/egusphere-2025-4716.pdf", "tier": "Tier 2", "interestedSource": "Yes", "relevantFinding": "Operational since 25 Feb 2025; improved precipitation and upper-air scores.", "limitations": "Preprint under discussion." }, { "id": "SRC-037", "role": "Supporting source", "title": "ECMWF's AI forecasts become operational", "authorOrOrganization": "European Centre for Medium-Range Weather Forecasts", "publicationDate": "2025-02-25", "url": "https://www.ecmwf.int/en/about/media-centre/news/2025/ecmwfs-ai-forecasts-become-operational", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "ECMWF placed its AI Forecasting System into operations and reported gains including up to 20% in tropical-cyclone track prediction.", "limitations": "Provider evaluation; accuracy gains differ by variable, lead time, and event." }, { "id": "SRC-103", "role": "Counterevidence", "title": "Forecast performance 2025", "authorOrOrganization": "European Centre for Medium-Range Weather Forecasts", "publicationDate": "2026-03-01", "url": "https://www.ecmwf.int/en/newsletter/187/news/forecast-performance-2025", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Reported gains vary by variable, lead time and event; forecast-skill improvements do not directly measure avoided human harm.", "limitations": "Provider evaluation, though operational and transparent." }, { "id": "SRC-R093", "role": "Counterevidence", "title": "Numerical models outperform AI weather forecasts of record-breaking extremes (Science Advances, 2026)", "authorOrOrganization": "Zhang, Fischer, Zscheischler, Engelke (ETH Zurich)", "publicationDate": "2026-04-29", "url": "https://arxiv.org/abs/2508.15724v1", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Physics-based HRES consistently beat GraphCast, Pangu-Weather and FuXi on record-breaking heat, cold and wind; AI models underestimated frequency and intensity.", "limitations": "Did not test ECMWF AIFS; published version DOI 10.1126/sciadv.aec1433." } ] }, { "id": "I-029", "area": "Science & discovery", "direction": "Gain", "claim": "A drug candidate whose target and molecule were both identified with generative AI (rentosertib, for idiopathic pulmonary fibrosis) entered a Phase III trial in July 2026 after a randomized Phase 2a trial.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Pipeline initiated by 2028", "geographicScope": "Global", "whoGains": "Analysis: people with idiopathic pulmonary fibrosis if the drug succeeds", "whoBearsTheCost": "Analysis: trial participants bearing investigational risk", "resolutionCriteria": "Pipeline claim graded on precursor only (R11): True if the Phase III trial remains active or reports by 2028-12-31 without termination for safety or futility; False if terminated for those reasons.", "ledgerEntry": null, "sources": [ { "id": "SRC-R035", "role": "Supporting source", "title": "Insilico announces initiation of Phase III trial for rentosertib", "authorOrOrganization": "Insilico Medicine", "publicationDate": "2026-07-07", "url": "https://insilico.com/news/xmjsn4l091", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Phase III initiated July 2026; target and molecule from the company's AI platforms; Phase 2a published in Nature Medicine.", "limitations": "Company disclosure." }, { "id": "SRC-R036", "role": "Supporting source", "title": "Insilico Medicine rentosertib enters Phase 3", "authorOrOrganization": "RuntimeWire", "publicationDate": null, "url": "https://runtimewire.com/article/insilico-medicine-rentosertib-phase-3-ai-drug-discovery", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Notes the first end-to-end AI-discovered claim is the company's and definition-dependent.", "limitations": "Secondary. Publication date not captured." }, { "id": "SRC-104", "role": "Counterevidence", "title": "ClinicalTrials.gov search records for rentosertib", "authorOrOrganization": "U.S. National Library of Medicine", "publicationDate": "2026-09-23", "url": "https://clinicaltrials.gov/search?term=rentosertib", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Registration documents pipeline stage and endpoints but does not establish Phase III efficacy, approval or comparative patient benefit.", "limitations": "Registry records are not trial results." } ] }, { "id": "I-030", "area": "Science & discovery", "direction": "Gain", "claim": "Evidence suggests AI protein-structure prediction (AlphaFold) is in routine use across structural biology, recognized by the 2024 Nobel Prize in Chemistry and used by researchers in about 190 countries.", "status": "Partial", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Structural biologists and drug-discovery researchers, especially groups without experimental structure capacity", "whoBearsTheCost": "Analysis: none identified", "resolutionCriteria": "True if an independent study by 2028-12-31 documents routine use of AI-predicted structures in published research; False if independent evidence shows marginal use or material error rates; Partially true otherwise.", "ledgerEntry": "By 2028-12-31, independent bibliometric evidence will confirm routine use of AI-predicted protein structures in published biomedical research.", "sources": [ { "id": "SRC-R037", "role": "Supporting source", "title": "The Nobel Prize in Chemistry 2024 (press release)", "authorOrOrganization": "Royal Swedish Academy of Sciences", "publicationDate": "2024-10-09", "url": "https://www.nobelprize.org/prizes/chemistry/2024/press-release/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Prize shared for computational protein design and protein structure prediction (AlphaFold); reports very wide research use.", "limitations": "Usage figure (more than two million people, 190 countries) verified against the Nobel press release 2026-09-23." }, { "id": "SRC-039", "role": "Supporting source", "title": "AlphaFold Protein Structure Database", "authorOrOrganization": "EMBL-EBI and Google DeepMind", "publicationDate": "2026-09-01", "url": "https://alphafold.ebi.ac.uk/", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Open database provides more than 200 million predicted protein structures.", "limitations": "Usage and availability do not establish downstream scientific or clinical outcomes." }, { "id": "SRC-105", "role": "Counterevidence", "title": "AlphaFold predictions are valuable hypotheses and accelerate but do not replace experimental structure determination", "authorOrOrganization": "Nature Methods", "publicationDate": "2023-11-30", "url": "https://www.nature.com/articles/s41592-023-02087-4", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Even very high-confidence predictions sometimes differ materially from experiment; routine use does not remove validation needs.", "limitations": "Selected comparisons, not an aggregate database evaluation." }, { "id": "SRC-R094", "role": "Counterevidence", "title": "AlphaFold predictions are valuable hypotheses and accelerate but do not replace experimental structure determination (Nature Methods 21:110-116)", "authorOrOrganization": "Terwilliger et al.", "publicationDate": "2023-11-30", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10776388/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Even very high-confidence AlphaFold predictions sometimes differed from experimental maps globally and locally; predictions should be treated as hypotheses.", "limitations": "Does not dispute availability, only reliability." } ] }, { "id": "I-031", "area": "Science & discovery", "direction": "Gain", "claim": "Evidence suggests AI coding agents have produced verifiable new results in algorithms and mathematics, such as Google DeepMind's AlphaEvolve finding a 4x4 complex matrix multiplication method using 48 scalar multiplications.", "status": "Partial", "currentReach": "Lab", "projectedReachByEOY2028": "Lab", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Researchers using discovered algorithms; Analysis: computing users if efficiency gains are deployed", "whoBearsTheCost": "Analysis: none identified", "resolutionCriteria": "True if at least 10 AI-originated results are independently verified by 2028-12-31; False if key reported results are shown incorrect or previously known; Partially true otherwise.", "ledgerEntry": "By 2028-12-31, at least 10 AI-originated mathematical or algorithmic results will be independently verified in peer-reviewed venues.", "sources": [ { "id": "SRC-R039", "role": "Supporting source", "title": "AlphaEvolve: a Gemini-powered coding agent for designing advanced algorithms", "authorOrOrganization": "Google DeepMind", "publicationDate": "2025-05-14", "url": "https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Reported a 4x4 complex matrix multiplication algorithm using 48 scalar multiplications, among other results.", "limitations": "Developer disclosure. URL and 48-multiplication result verified 2026-09-23." }, { "id": "SRC-107", "role": "Counterevidence", "title": "Artificial intelligence and illusions of understanding in scientific research", "authorOrOrganization": "Nature", "publicationDate": "2024-03-06", "url": "https://www.nature.com/articles/s41586-024-07146-0", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "AI can increase output while narrowing approaches and creating false confidence; verifiable examples do not establish broad autonomous discovery.", "limitations": "Conceptual analysis rather than a direct replication." } ] }, { "id": "I-032", "area": "Science & discovery", "direction": "Gain", "claim": "Evidence suggests AI co-scientist systems can produce hypotheses that survive laboratory testing in selected biomedical domains.", "status": "Partial", "currentReach": "Lab", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Scientists, patients, and research funders", "whoBearsTheCost": "Researchers diverted by plausible but weak hypotheses", "resolutionCriteria": "By 2028, independent prospective comparisons establish whether hit rate exceeds expert/search baselines.", "ledgerEntry": "AI co-scientists beat expert/search baselines on prospective hit rate in two domains by EOY 2028.", "sources": [ { "id": "SRC-041", "role": "Supporting source", "title": "An AI co-scientist for accelerating scientific discovery", "authorOrOrganization": "Nature", "publicationDate": "2026-06-01", "url": "https://www.nature.com/articles/s41586-026-10644-y", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "System-generated hypotheses were experimentally evaluated in selected biomedical settings and included validated findings.", "limitations": "Developer-led benchmarks and selected domains; replication and comparative hit rates remain limited." }, { "id": "SRC-R095", "role": "Counterevidence", "title": "An AI Co-Scientist Aced This Test (IEEE Spectrum)", "authorOrOrganization": "IEEE Spectrum", "publicationDate": null, "url": "https://spectrum.ieee.org/amp/ai-co-scientist-2674000043", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "The headline validated result recapitulated an Imperial team's unpublished, already-obtained finding rather than producing a new discovery.", "limitations": "Journalism; supportive in tone. Date not captured." }, { "id": "R66-E18-1", "role": "Supporting source", "title": "ART enzyme-system research", "authorOrOrganization": "Anthropic", "publicationDate": "2026-09-23", "url": "https://www.anthropic.com/news/claude-discovers-novel-enzyme-system", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Reports roughly 950 agents scanning genomic data, selecting an ART candidate system and initiating a biological-discovery workflow.", "limitations": "Function remains unknown. Humans performed all wet-lab work. Previously described reverse transcriptase underlies the finding. No proven new gene-editing therapy or autonomous wet-lab control." }, { "id": "R66-E19-1", "role": "Supporting source", "title": "ART novelty and validation critique", "authorOrOrganization": "WIRED", "publicationDate": "2026-09-29", "url": "https://www.wired.com/story/anthropic-says-it-discovered-a-crispr-like-system-now-what", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Reports expert disagreement over novelty and emphasizes need to establish function experimentally.", "limitations": "Expert criticism is not a completed replication or disproof. It narrows the discovery claim rather than denying the usefulness of AI analysis." } ], "lastReviewed": "2026-10-06" }, { "id": "I-033", "area": "Science & discovery", "direction": "Gain", "claim": "Evidence suggests an autonomous AI-guided laboratory can run closed-loop inorganic synthesis at high throughput, realizing 36 of 57 target compounds over 17 days of continuous operation.", "status": "Partial", "currentReach": "Lab", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Materials researchers and downstream clean-energy or manufacturing users", "whoBearsTheCost": "Labs facing high capital costs or brittle automation", "resolutionCriteria": "By 2028, independent benchmarked labs reproduce a >=25% cycle reduction.", "ledgerEntry": "Autonomous materials labs cut prespecified optimization cycles >=25% by EOY 2028.", "sources": [ { "id": "SRC-043", "role": "Supporting source", "title": "An autonomous laboratory for the accelerated synthesis of novel materials", "authorOrOrganization": "Nature", "publicationDate": "2023-11-29", "url": "https://www.nature.com/articles/s41586-023-06734-w", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Autonomous laboratory combined literature, ML, robotics, and synthesis to produce target inorganic materials.", "limitations": "Specialized lab; novelty and generality were disputed." }, { "id": "SRC-R096", "role": "Counterevidence", "title": "Challenges in High-Throughput Inorganic Materials Prediction and Autonomous Synthesis (PRX Energy 3, 011002)", "authorOrOrganization": "Leeman, Liu, Stiles, Lee, Bhatt, Schoop, Palgrave", "publicationDate": "2024-03-07", "url": "https://par.nsf.gov/servlets/purl/10532478", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Re-analysis of the A-Lab's 43 claimed new materials concluded none were new; about two-thirds likely known disordered versions.", "limitations": "A-Lab paper later corrected (now 36 compounds from 57 targets, 'inorganic materials')." } ] }, { "id": "I-034", "area": "Science & discovery", "direction": "Harm", "claim": "Evidence suggests AI materials-discovery systems can overstate novelty when training and reference databases overlap.", "status": "Partial", "currentReach": "Lab", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O2", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Researchers who adopt stronger novelty audits", "whoBearsTheCost": "Funders and scientists pursuing rediscoveries", "resolutionCriteria": "By 2028, preregistered novelty audits across >=3 systems quantify overlap and practical novelty.", "ledgerEntry": "At least one additional AI materials system has >=25% claimed-novel database overlap by EOY 2028.", "sources": [ { "id": "SRC-045", "role": "Supporting source", "title": "On the novelty of autonomous materials discovery", "authorOrOrganization": "Scientific Reports", "publicationDate": "2025-06-01", "url": "https://www.nature.com/articles/s41598-025-10672-0", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Reanalysis found 42 of 43 reported crystals already represented in the Materials Project, challenging novelty claims.", "limitations": "Database membership and practical synthesis novelty are not identical." }, { "id": "SRC-R097", "role": "Counterevidence", "title": "Scaling deep learning for materials discovery (Nature 624)", "authorOrOrganization": "Merchant et al. (Google DeepMind)", "publicationDate": "2023-11-29", "url": "https://doi.org/10.1038/s41586-023-06735-9", "tier": "Tier 1", "interestedSource": "Yes", "relevantFinding": "Developer reports 736 predicted structures independently realized experimentally by external groups.", "limitations": "Developer-authored (interested). Figure verified 2026-09-23." }, { "id": "R66-E18-1", "role": "Counterevidence", "title": "ART enzyme-system research", "authorOrOrganization": "Anthropic", "publicationDate": "2026-09-23", "url": "https://www.anthropic.com/news/claude-discovers-novel-enzyme-system", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Reports roughly 950 agents scanning genomic data, selecting an ART candidate system and initiating a biological-discovery workflow.", "limitations": "Function remains unknown. Humans performed all wet-lab work. Previously described reverse transcriptase underlies the finding. No proven new gene-editing therapy or autonomous wet-lab control." }, { "id": "R66-E19-1", "role": "Counterevidence", "title": "ART novelty and validation critique", "authorOrOrganization": "WIRED", "publicationDate": "2026-09-29", "url": "https://www.wired.com/story/anthropic-says-it-discovered-a-crispr-like-system-now-what", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Reports expert disagreement over novelty and emphasizes need to establish function experimentally.", "limitations": "Expert criticism is not a completed replication or disproof. It narrows the discovery claim rather than denying the usefulness of AI analysis." } ], "lastReviewed": "2026-10-06" }, { "id": "I-035", "area": "Science & discovery", "direction": "Harm", "claim": "Evidence suggests LLM-generated clinical literature reviews can be less comprehensive and contain less accurate citations than human reviews.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Reviewers who use audited tools for retrieval support", "whoBearsTheCost": "Clinicians, patients, and researchers relying on incorrect syntheses", "resolutionCriteria": "By 2028, independent benchmarks establish whether current systems meet non-inferiority thresholds.", "ledgerEntry": "LLM clinical reviews remain >=10 points worse on recall or citation accuracy by EOY 2028.", "sources": [ { "id": "SRC-R040", "role": "Supporting source", "title": "Delving into LLM-assisted writing in biomedical publications through excess vocabulary (Science Advances, 2025)", "authorOrOrganization": "Kobak, Gonzalez-Marquez, Horvat, Lause", "publicationDate": "2025-07-02", "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12219543/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "At least 13.5% of 2024 PubMed abstracts processed with LLMs, estimated from excess style words.", "limitations": "Lower-bound estimate from excess style words; measures LLM processing, not errors. DOI 10.1126/sciadv.adt3813, date and 13.5% figure verified 2026-09-23." }, { "id": "SRC-047", "role": "Supporting source", "title": "Large language model generated clinical reviews are less comprehensive and contain less accurate citations", "authorOrOrganization": "npj Digital Medicine", "publicationDate": "2025-03-01", "url": "https://www.nature.com/articles/s41746-025-01535-z", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Comparison found LLM-generated clinical reviews less comprehensive with less accurate citations than human-authored reviews.", "limitations": "Models and prompting evolve; one review task does not cover all scientific uses." }, { "id": "SRC-108", "role": "Counterevidence", "title": "Large language model generated clinical reviews are less comprehensive and contain less accurate citations", "authorOrOrganization": "npj Digital Medicine", "publicationDate": "2025-03-01", "url": "https://www.nature.com/articles/s41746-025-01535-z", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Direct benchmarking finds lower comprehensiveness and citation accuracy, but does not establish that AI-assisted human writing degrades the literature overall.", "limitations": "One review task and rapidly changing models." }, { "id": "SRC-R098", "role": "Counterevidence", "title": "LLMs for abstract screening in Cochrane systematic reviews (JAMIA 32:893-904)", "authorOrOrganization": "Oxford researchers", "publicationDate": "2025-05-01", "url": "https://www.oncology.ox.ac.uk/publication/2093498", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Six LLMs across 23 Cochrane reviews matched or exceeded human sensitivity in abstract screening, with workload reductions of 38% to 99% in LLM-human ensembles.", "limitations": "Screening only, not full review writing; precision drops at scale." } ] }, { "id": "I-036", "area": "Cost of living & productivity", "direction": "Harm", "claim": "Data-center load growth, much of it for AI, could keep raising electricity costs for households in the PJM region; the grid's market monitor attributes $29.4 billion in capacity charges over four auctions to data-center load, but does not isolate AI's share.", "status": "Watch", "currentReach": "Population-scale", "projectedReachByEOY2028": "Population-scale", "outcomeEvidence": "O4", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Sub-national", "whoGains": "Generators receiving capacity payments; data-center operators whose load drove the costs", "whoBearsTheCost": "Residential and business electricity customers across PJM", "resolutionCriteria": "True if by 2028-12-31 the data-center share of capacity charges is at or above 40% and at least one source isolates AI's share of data-center load; False if the share falls below 20%; Partially true if the share stays high but AI's share is not isolated.", "ledgerEntry": "By 2028-12-31, data-center load will account for at least 40% of PJM capacity charges, with AI's share of that load isolated by at least one independent source.", "sources": [ { "id": "SRC-R041", "role": "Supporting source", "title": "PJM market monitor annual report: clear warning signs", "authorOrOrganization": "Utility Dive", "publicationDate": null, "url": "https://www.utilitydive.com/news/pjm-capacity-energy-market-reliability-monitoring-analytics/814647/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Market monitor: data-center load added $23.1B over three auctions; wholesale power cost $67B in 2025, up 54%.", "limitations": "Journalism summarizing the monitor; publication date not captured." }, { "id": "SRC-R042", "role": "Supporting source", "title": "Market Monitor Report, IMMMC Information Webinar", "authorOrOrganization": "Monitoring Analytics (PJM independent market monitor)", "publicationDate": "2026-07-27", "url": "https://www.pjm.com/-/media/DotCom/committees-groups/committees/mc/2026/20260727-web/item-02---marketing-monitoring-report---presentation.pdf", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "2028/29 auction: $6.3B of $16.4B (38.2%) from data-center load; $29.4B (46.2%) over four auctions.", "limitations": "Data centers, not AI specifically." }, { "id": "SRC-053", "role": "Supporting source", "title": "PJM customers paid billions for data-center transmission costs", "authorOrOrganization": "Utility Dive", "publicationDate": "2025-09-15", "url": "https://www.utilitydive.com/news/pjm-data-center-transmission-costs-ratepayers/761579/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Reporting based on regulatory records estimated $4.4 billion in data-centre-related transmission costs paid by customers in seven PJM states.", "limitations": "Allocation methods and counterfactual system investment are contested; not all load is AI." }, { "id": "SRC-109", "role": "Counterevidence", "title": "DOE report on electricity demand from data centers", "authorOrOrganization": "U.S. Department of Energy", "publicationDate": "2024-12-20", "url": "https://www.energy.gov/articles/doe-releases-new-report-evaluating-increase-electricity-demand-data-centers", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Projected data-center electricity growth does not isolate AI load or determine who ultimately bears grid costs.", "limitations": "Scenario range is broad and includes non-AI computing." }, { "id": "SRC-R101", "role": "Counterevidence", "title": "Retail Electricity Price Trends and Drivers: Data Update, 2026 Edition", "authorOrOrganization": "Lawrence Berkeley National Laboratory and The Brattle Group", "publicationDate": "2026-03-01", "url": "https://eta-publications.lbl.gov/sites/default/files/2026-03/retail_price_trends_2026_edition.pdf", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "State-level load growth was linked to declining average retail prices 2019 to 2025 in most states; PJM capacity prices are the notable exception.", "limitations": "Relationship weaker for residential prices; past pattern may not persist." }, { "id": "R66-E25-1", "role": "Context / limits", "title": "DOE SPARK grid projects", "authorOrOrganization": "US Department of Energy", "publicationDate": "2026-09-24", "url": "https://www.energy.gov/articles/energy-department-announces-speed-power-investments-across-26-states-lower-electricity", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Announces intended support for 31 projects in 26 states to unlock more than 23 GW of capacity.", "limitations": "Total project value differs from federal funding. Planned capacity is not current operation or proof of household bill increases caused specifically by AI." } ], "lastReviewed": "2026-10-06" }, { "id": "I-037", "area": "Cost of living & productivity", "direction": "Harm", "claim": "Evidence suggests AI data-center demand for memory has driven record DRAM and flash price increases in 2026, raising component costs for consumer PCs and smartphones.", "status": "Partial", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O2", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Memory manufacturers; cloud providers securing long-term supply", "whoBearsTheCost": "Consumers and device makers buying PCs, laptops and smartphones", "resolutionCriteria": "True if an independent price index shows consumer PC or smartphone prices rose in 2026 to 2027 with memory costs cited as a driver; False if consumer device prices fell over the period; Partially true if only component prices rose.", "ledgerEntry": "By 2028-12-31, independent price data will show AI-driven memory costs raised consumer PC or smartphone prices during 2026 to 2027.", "sources": [ { "id": "SRC-R043", "role": "Supporting source", "title": "TrendForce Upgrades Memory Price Outlook for 1Q 2026", "authorOrOrganization": "TrendForce (via EE Times Asia)", "publicationDate": "2026-02-02", "url": "https://www.eetasia.com/?p=72392", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Conventional DRAM contract prices revised to +90 to 95% QoQ; PC DRAM over 100%; driven by AI and data-center demand plus strong Q4 2025 PC shipments.", "limitations": "Forecast, not realized retail prices." }, { "id": "SRC-R044", "role": "Supporting source", "title": "DRAM prices expected to double in Q1 as AI ambitions push memory fabs to their limit", "authorOrOrganization": "The Register", "publicationDate": "2026-02-02", "url": "https://www.theregister.com/2026/02/02/dram_prices_expected_to_double/", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "AI-driven hyperscalers and cloud providers largely blamed; OEM system prices expected to rise as inventories deplete.", "limitations": "Secondary journalism." }, { "id": "SRC-R068", "role": "Supporting source", "title": "DRAM and NAND remain more expensive: TrendForce sees slowing but still rising memory prices in Q3 2026", "authorOrOrganization": "Igor's Lab (reporting TrendForce)", "publicationDate": null, "url": "https://www.igorslab.de/en/dram-and-nand-remain-more-expensive-trendforce-sees-slowing-but-still-rising-memory-prices-q3-2026/", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Conventional DRAM forecast +13 to 18% QoQ in Q3 2026; notebook makers passing costs into prices.", "limitations": "Secondary; date not captured." }, { "id": "SRC-110", "role": "Counterevidence", "title": "Smaller phone and laptop makers dig in for years of memory scarcity", "authorOrOrganization": "Reuters", "publicationDate": "2026-09-16", "url": "https://www.reuters.com/business/retail-consumer/smaller-phone-laptop-makers-dig-years-memory-scarcity-2026-09-16/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Manufacturers describe a broader capacity crunch and uneven price pass-through; AI demand is a driver but its marginal share is not isolated.", "limitations": "Journalistic synthesis relying partly on industry forecasts." } ] }, { "id": "I-038", "area": "Cost of living & productivity", "direction": "Gain", "claim": "Evidence suggests US adults' consumer surplus from generative AI tools rose from about $116 billion (July 2025) to $172 billion (March 2026), far exceeding the revenue these tools earn.", "status": "Partial", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Frequent users, workplace users and paid subscribers, who report the highest valuations", "whoBearsTheCost": "Analysis: none directly; costs of the infrastructure are borne elsewhere (see CLAIM-025, CLAIM-026)", "resolutionCriteria": "True if the tracker or an independent replication shows at least $250 billion by 2028-12-31; False if at or below $116 billion; Partially true if between.", "ledgerEntry": "By 2028-12-31, estimated US consumer surplus from generative AI will reach at least $250 billion.", "sources": [ { "id": "SRC-R045", "role": "Supporting source", "title": "What is Generative AI Worth?", "authorOrOrganization": "Brynjolfsson, Collis, Eggers, Kazinnik, Nguyen", "publicationDate": "2026-04-13", "url": "https://digitaleconomy.stanford.edu/app/uploads/2026/04/WhatsGenAIWorth.pdf", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Mean WTA $98 (2025) to $124.50 (2026); users 98M to 115M; consumer surplus $116B to $172B.", "limitations": "Working paper; online choice experiments." }, { "id": "SRC-R071", "role": "Supporting source", "title": "GDP-B surplus observer: What is generative AI worth?", "authorOrOrganization": "Stanford Digital Economy Lab", "publicationDate": null, "url": "https://digitaleconomy.stanford.edu/project/indicators/gdp-b-surplus-observer/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Tracker shows $116B (July 2025) and $183B (July 2026).", "limitations": "Tracker page; date not captured." }, { "id": "SRC-111", "role": "Counterevidence", "title": "Hypothetical bias in stated choice experiments: integrative synthesis", "authorOrOrganization": "Independent academic investigators", "publicationDate": "2021-02-01", "url": "https://arxiv.org/abs/2102.02940", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Stated-choice valuations can show material, context-dependent hypothetical bias, limiting national consumer-surplus extrapolation.", "limitations": "Methodological review, not a GenAI replication." } ] }, { "id": "I-039", "area": "Cost of living & productivity", "direction": "Harm", "claim": "Algorithmic rent setting software could raise rents where widely adopted; a White House analysis estimated it added $3.8 billion to US rents in 2023, about $70 a month per affected unit, but AI's specific role is not established.", "status": "Watch", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O2", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Landlords using the software", "whoBearsTheCost": "Renters in buildings using algorithmic pricing, especially in high-adoption metros", "resolutionCriteria": "True if by 2028-12-31 an independent study finds a significant rent premium attributable to AI-based pricing; False if no premium; Partially true if a premium is found for algorithmic pricing without AI attribution.", "ledgerEntry": "By 2028-12-31, an independent study will attribute a significant rent premium to AI-based rent-pricing software.", "sources": [ { "id": "SRC-R046", "role": "Supporting source", "title": "RealPage DOJ settlement", "authorOrOrganization": "HousingWire", "publicationDate": null, "url": "https://www.housingwire.com/articles/realpage-doj-settlement/", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Settlement without penalty restricts nonpublic data use and training on active lease data.", "limitations": "Publication date not captured (Nov 2025)." }, { "id": "SRC-R047", "role": "Supporting source", "title": "The Cost of Anticompetitive Pricing Algorithms in Rental Housing", "authorOrOrganization": "White House Council of Economic Advisers (archived)", "publicationDate": "2024-12-17", "url": "https://bidenwhitehouse.archives.gov/cea/written-materials/2024/12/17/the-cost-of-anticompetitive-pricing-algorithms-in-rental-housing", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Algorithmic pricing cost renters in algorithm-using buildings about $70 a month; $3.8B total in 2023 (likely lower bound).", "limitations": "Company disputes assumptions; algorithm type not isolated." }, { "id": "SRC-R069", "role": "Supporting source", "title": "Pricing software raises Houston-area rent, report finds", "authorOrOrganization": "Axios Houston", "publicationDate": "2025-01-06", "url": "https://www.axios.com/local/houston/2025/01/06/realpage-rent-prices-doj-lawsuit", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "About 40% of Houston landlords and nearly 70% in Atlanta used the software; RealPage controls about 80% of rent setting software.", "limitations": "Secondary reporting of CEA analysis." }, { "id": "SRC-112", "role": "Counterevidence", "title": "Justice Department sues RealPage for algorithmic pricing scheme", "authorOrOrganization": "U.S. Department of Justice", "publicationDate": "2024-08-23", "url": "https://www.justice.gov/archives/opa/pr/justice-department-sues-realpage-algorithmic-pricing-scheme-harms-millions-american", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "The case concerns alleged information sharing and algorithmic coordination; it does not isolate generative AI or prove the CEA rent estimate.", "limitations": "Complaint allegations are not final adjudicated findings." } ] }, { "id": "I-040", "area": "Cost of living & productivity", "direction": "Harm", "claim": "Algorithmic and AI-driven personalized pricing could raise the prices some consumers pay for the same goods; the FTC found retailers' pricing intermediaries use personal data such as location, browsing and cart behavior to set individualized prices, but net consumer effects are not yet measured.", "status": "Watch", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O0", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Retailers and pricing intermediaries; Analysis: some price-sensitive consumers offered discounts", "whoBearsTheCost": "Not yet identifiable; Analysis: consumers assessed as willing to pay more", "resolutionCriteria": "True if by 2028-12-31 an independent study or FTC finding shows AI-based personalized pricing raised prices for identifiable consumer groups by 5% or more; False if it finds no net increase; Partially true if increases are found without AI attribution.", "ledgerEntry": "By 2028-12-31, an independent study or FTC finding will show AI-based personalized pricing raised prices for identifiable US consumer groups by at least 5%.", "sources": [ { "id": "SRC-R073", "role": "Supporting source", "title": "FTC Surveillance Pricing 6(b) Study: Research Summaries, A Staff Perspective", "authorOrOrganization": "Federal Trade Commission staff", "publicationDate": "2025-01-17", "url": "https://ftc.gov/system/files/ftc_gov/pdf/p246202_surveillancepricing6bstudy_researchsummaries_redacted.pdf", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Intermediaries enable retailers to set individualized prices using data such as precise location, demographics, browsing and cart behavior.", "limitations": "Preliminary staff observations; based on documents from six intermediaries; no measured consumer price effects." }, { "id": "SRC-R074", "role": "Supporting source", "title": "House Energy and Commerce Democrats: surveillance pricing inquiry letter", "authorOrOrganization": "House Energy and Commerce Committee (Democratic staff)", "publicationDate": "2026-05-11", "url": "https://democrats-energycommerce.house.gov/sites/evo-subsites/democrats-energycommerce.house.gov/files/evo-media-document/surveillance-pricing-inquiry-5.11.26.pdf", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Notes the FTC opened an investigation into a grocery-delivery company's AI pricing tool in Dec 2025, after randomized price tests drew criticism.", "limitations": "Partisan document; the investigation itself was not independently verified." }, { "id": "SRC-R075", "role": "Supporting source", "title": "FTC proposes enforcement policy statement on personalized pricing", "authorOrOrganization": "Holland & Knight", "publicationDate": "2026-08-21", "url": "https://www.hklaw.com/en/insights/publications/2026/08/ftc-proposes-enforcement-policy-statement-on-personalized-pricing", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "FTC sought comment (Aug 19, 2026) on a policy statement on personalized pricing; cites research that such pricing likely raises profits with losses to some consumers.", "limitations": "Law-firm summary; primary FTC statement not fetched." }, { "id": "SRC-R076", "role": "Supporting source", "title": "ABA Antitrust Panel Recap: Be on the Lookout, Surveillance Pricing", "authorOrOrganization": "American Bar Association", "publicationDate": null, "url": "https://www.americanbar.org/groups/antitrust_law/resources/newsletters/panel-recap-be-on-lookout/", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "New York law effective July 2025 requires a disclosure when prices are set by an algorithm using personal data.", "limitations": "Secondary; publication date not captured." }, { "id": "SRC-125", "role": "Counterevidence", "title": "Surveillance Pricing 6(b) Study: Staff Perspective", "authorOrOrganization": "Federal Trade Commission", "publicationDate": "2026-01-01", "url": "https://www.ftc.gov/system/files/ftc_gov/pdf/p246202_surveillancepricing6bstudy_researchsummaries_redacted.pdf", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "The inquiry documents data and pricing practices but does not measure net price effects or isolate AI's contribution.", "limitations": "Preliminary staff observations from six intermediaries." } ] }, { "id": "I-041", "area": "Cost of living & productivity", "direction": "Gain", "claim": "AI-adopting firms may raise real sales per employee by at least 1% by 2028.", "status": "Watch", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O0", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Workers, owners, and customers if gains pass through", "whoBearsTheCost": "Non-adopting firms and workers if gains concentrate", "resolutionCriteria": "By 2028, representative administrative data with credible identification confirms or rejects a >=1% gain.", "ledgerEntry": "AI-adopting firms achieve >=1% higher real sales per employee by EOY 2028.", "sources": [ { "id": "SRC-051", "role": "Supporting source", "title": "Global evidence on business use of AI", "authorOrOrganization": "NBER Digest", "publicationDate": "2026-05-01", "url": "https://www.nber.org/digest/202605/global-evidence-business-use-ai", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Business surveys imply expected gains in productivity and output, but current realized effects remain limited.", "limitations": "Expectations are not realized causal impacts." }, { "id": "SRC-R100", "role": "Counterevidence", "title": "The Simple Macroeconomics of AI (NBER WP 32487)", "authorOrOrganization": "Acemoglu", "publicationDate": "2024-05-01", "url": "https://www.nber.org/papers/w32487", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Task-based estimate implies no more than a 0.66% total factor productivity gain over 10 years, possibly under 0.53%.", "limitations": "Model-based projection. URL and figures verified 2026-09-23." } ] }, { "id": "I-042", "area": "Cost of living & productivity", "direction": "Harm", "claim": "U.S. data centres may reach at least 6.7% of national electricity demand by 2028, increasing pressure on generation and grids.", "status": "Watch", "currentReach": "Population-scale", "projectedReachByEOY2028": "Population-scale", "outcomeEvidence": "O0", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Cloud and AI users and host communities", "whoBearsTheCost": "Ratepayers and communities facing constrained power or infrastructure", "resolutionCriteria": "Official 2028 electricity accounts establish the share and associated reliability/cost effects.", "ledgerEntry": "U.S. data centres use >=6.7% of national electricity in 2028 with documented system pressure.", "sources": [ { "id": "SRC-055", "role": "Supporting source", "title": "DOE releases report evaluating increase in electricity demand from data centers", "authorOrOrganization": "U.S. Department of Energy", "publicationDate": "2024-12-20", "url": "https://www.energy.gov/articles/doe-releases-new-report-evaluating-increase-electricity-demand-data-centers", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Data centres used about 4.4% of U.S. electricity in 2023 and are projected at 6.7% to 12% by 2028.", "limitations": "Projection covers all data-centre demand, not AI alone." }, { "id": "R66-E25-1", "role": "Context / limits", "title": "DOE SPARK grid projects", "authorOrOrganization": "US Department of Energy", "publicationDate": "2026-09-24", "url": "https://www.energy.gov/articles/energy-department-announces-speed-power-investments-across-26-states-lower-electricity", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Announces intended support for 31 projects in 26 states to unlock more than 23 GW of capacity.", "limitations": "Total project value differs from federal funding. Planned capacity is not current operation or proof of household bill increases caused specifically by AI." } ], "lastReviewed": "2026-10-06" }, { "id": "I-043", "area": "Cost of living & productivity", "direction": "Gain", "claim": "AI-driven productivity gains may reduce quality-adjusted consumer prices by at least 0.5% in two service sectors by 2028.", "status": "Watch", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O0", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Multi-country", "whoGains": "Consumers and competitive service providers", "whoBearsTheCost": "Workers or smaller firms if savings come from displacement", "resolutionCriteria": "By 2028, official or independent price studies identify two sectors meeting the threshold.", "ledgerEntry": "AI lowers quality-adjusted consumer prices >=0.5% in two service sectors by EOY 2028.", "sources": [ { "id": "SRC-057", "role": "Supporting source", "title": "The effects of generative AI on productivity, innovation and entrepreneurship", "authorOrOrganization": "OECD", "publicationDate": "2025-11-01", "url": "https://www.oecd.org/en/publications/the-effects-of-generative-ai-on-productivity-innovation-and-entrepreneurship_b21df222-en.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Review finds meaningful task-level gains but strong heterogeneity by task, worker, implementation, and measurement horizon.", "limitations": "Synthesis spans unlike studies; firm and economy-wide pass-through remains uncertain." }, { "id": "SRC-R102", "role": "Counterevidence", "title": "TrendForce Upgrades Memory Price Outlook for 1Q 2026", "authorOrOrganization": "TrendForce (via EE Times Asia)", "publicationDate": "2026-02-02", "url": "https://www.eetasia.com/?p=72392", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "AI and data-center demand drove record DRAM contract price increases, pushing up component costs for consumer devices.", "limitations": "Component prices, not retail pass-through." } ] }, { "id": "I-044", "area": "Cost of living & productivity", "direction": "Harm", "claim": "Small and medium-sized firms could fall behind large firms in realized AI productivity benefits; current evidence shows an adoption gap, not a measured benefit gap.", "status": "Watch", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O1", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Multi-country", "whoGains": "Large firms with data, skills, and integration budgets", "whoBearsTheCost": "SME owners, workers, and communities", "resolutionCriteria": "By 2028, representative business data test both adoption and realized productivity gaps.", "ledgerEntry": "SMEs trail large firms by >=20 points in AI adoption and >=1% productivity growth by EOY 2028.", "sources": [ { "id": "SRC-059", "role": "Supporting source", "title": "Generative AI and the SME workforce", "authorOrOrganization": "OECD", "publicationDate": "2025-11-01", "url": "https://www.oecd.org/en/publications/generative-ai-and-the-sme-workforce_2d08b99d-en/full-report/component-4.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "SME evidence shows adoption and benefits are uneven, with skills, data, cost, and management constraints.", "limitations": "Surveys and case studies; causal productivity measurement remains thin." }, { "id": "SRC-R103", "role": "Counterevidence", "title": "Is AI Use Increasing Among Small Businesses?", "authorOrOrganization": "U.S. Census Bureau", "publicationDate": "2024-12-01", "url": "https://census.gov/newsroom/blogs/research-matters/2024/12/ai-use-small-businesses.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Business Trends and Outlook Survey data show AI use accelerating among small firms; smallest firms had one of the highest use rates in the period examined.", "limitations": "2024 data; 2026 Census data show largest firms about double the smallest." } ] }, { "id": "I-045", "area": "Independence & accessibility", "direction": "Gain", "claim": "An AI-decoded speech neuroprosthesis has restored communication for a man with ALS, decoding his intended speech with about 97.5% word accuracy.", "status": "Confirmed", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Sub-national", "whoGains": "People with ALS or paralysis who have lost speech", "whoBearsTheCost": "Analysis: trial participants bearing surgical risk", "resolutionCriteria": "True if by 2028-12-31 at least 10 people use AI speech neuroprostheses in trials with accuracy of 90% or more; False if trials are halted for safety; Partially true otherwise.", "ledgerEntry": "By 2028-12-31, at least 10 people with severe speech loss will communicate daily through AI speech neuroprostheses in clinical trials.", "sources": [ { "id": "SRC-R049", "role": "Supporting source", "title": "An Accurate and Rapidly Calibrating Speech Neuroprosthesis (NEJM, 2024)", "authorOrOrganization": "Card et al. (UC Davis)", "publicationDate": "2024-08-14", "url": "https://pubmed.ncbi.nlm.nih.gov/39141853/", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Participant with ALS communicated via decoded speech at about 97.5% accuracy.", "limitations": "Single participant (BrainGate2, NCT00912041). DOI 10.1056/NEJMoa2314132, date and 97.5% figure verified 2026-09-23." }, { "id": "SRC-114", "role": "Counterevidence", "title": "Artificial intelligence for brain-to-speech decoding in paralysis: a systematic review", "authorOrOrganization": "Independent academic investigators", "publicationDate": "2026-05-01", "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13383455/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "The review finds pervasive patient-selection bias and predominantly tiny, short studies, limiting generalization from one ALS participant.", "limitations": "Rapidly evolving and heterogeneous field." } ] }, { "id": "I-046", "area": "Independence & accessibility", "direction": "Mixed", "claim": "Evidence suggests multimodal AI visual assistants give blind and low-vision people more independent access to visual information, but that independence depends on sharing images of private spaces and documents with AI providers.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O1", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Blind and low-vision users of AI visual assistants", "whoBearsTheCost": "The same users, through privacy exposure of images shared with providers", "resolutionCriteria": "See MIXED COMPONENTS for component-level criteria.", "ledgerEntry": "By 2028-12-31, quantitative evidence will confirm improved independence for BLV users of AI visual assistants alongside documented privacy trade-offs.", "sources": [ { "id": "SRC-R050", "role": "Supporting source", "title": "How Multimodal Large Language Models Support Access to Visual Information: A Diary Study With Blind and Low Vision People (CHI 2026)", "authorOrOrganization": "Gonzalez Penuela, Jung, Lin, Hu, Azenkot (Cornell)", "publicationDate": "2026-02-13", "url": "https://arxiv.org/abs/2602.13469", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Diary study of BLV people using MLLM visual assistants in daily life; accuracy and goal-directed assistance both matter.", "limitations": "Qualitative; small sample." }, { "id": "SRC-R051", "role": "Supporting source", "title": "Before, I Asked My Mom, Now I Ask ChatGPT: Visual Privacy Management with Generative AI for Blind and Low-Vision People", "authorOrOrganization": "arXiv preprint (authors not captured)", "publicationDate": null, "url": "https://arxiv.org/pdf/2507.00286", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "BLV users shift visual questions from family to generative AI, raising visual privacy management issues.", "limitations": "Preprint; details and venue not verified; date approximate (Jul 2025)." }, { "id": "SRC-115", "role": "Counterevidence", "title": "AI accessibility performance gaps for disabled users: a review", "authorOrOrganization": "PubMed Central review", "publicationDate": "2026-03-01", "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12980396/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "AI systems often perform worse for disabled users across speech, vision and interface domains, limiting generalization from small prototypes.", "limitations": "Heterogeneous systems and evaluations." }, { "id": "SRC-127", "role": "Counterevidence", "title": "Beyond Blanket Masking: Examining Granularity for Privacy Protection in Images Captured by Blind and Low Vision Users", "authorOrOrganization": "Murrugarra-Llerena et al. (Stony Brook University, UT Austin, University of Maryland)", "publicationDate": "2025-08-12", "url": "https://arxiv.org/abs/2508.09245", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "A privacy-protection framework selectively masked high-risk information in images captured by blind and low-vision users while preserving more useful image content, showing that visual-assistance utility need not require exposing all sensitive content.", "limitations": "Prototype and benchmark evidence, not a deployed service or longitudinal study with blind and low-vision users; it does not establish provider retention practices or real-world incident rates." } ] }, { "id": "I-047", "area": "Independence & accessibility", "direction": "Harm", "claim": "Evidence suggests commercial speech recognition systems tested in 2019 made nearly twice as many errors for Black speakers as for white speakers (average word error rate 0.35 vs. 0.19 across five systems), which could disadvantage those users in captioning and voice-driven services; current systems have not been re-audited on the same basis.", "status": "Partial", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Analysis: speakers whose speech the systems recognize well", "whoBearsTheCost": "Black speakers, particularly speakers of African American English, relying on speech recognition", "resolutionCriteria": "True if an independent audit by 2028-12-31 finds a gap of 0.10 or more; False if below 0.03; Partially true if between.", "ledgerEntry": "By 2028-12-31, independent audits will still find a word-error-rate gap of at least 0.10 between Black and white speakers on major speech-recognition systems.", "sources": [ { "id": "SRC-R052", "role": "Supporting source", "title": "Racial disparities in automated speech recognition (PNAS, 2020)", "authorOrOrganization": "Koenecke et al.", "publicationDate": "2020-03-23", "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC7149386", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Five commercial systems: average word error rate 0.35 for Black speakers vs. 0.19 for white speakers.", "limitations": "2019 systems. DOI 10.1073/pnas.1915768117 and figures verified 2026-09-23." }, { "id": "SRC-116", "role": "Counterevidence", "title": "Reexamining Racial Disparities in Automatic Speech Recognition Performance: Confounding by Provenance", "authorOrOrganization": "Independent academic investigators", "publicationDate": "2024-07-19", "url": "https://arxiv.org/abs/2407.13982", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Recording provenance materially confounds the racial-gap estimate in the benchmark used by the original study.", "limitations": "Preprint; confounding does not prove the disparity is absent." } ] }, { "id": "I-048", "area": "Independence & accessibility", "direction": "Harm", "claim": "The FTC found that an AI-powered web-accessibility plug-in did not make websites compliant with accessibility guidelines as advertised, leaving disabled users facing barriers on sites marketed as accessible; the company paid $1 million under a final order in April 2025.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O1", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Analysis: businesses seeking low-cost compliance (short term)", "whoBearsTheCost": "Blind, deaf and other disabled users of affected websites", "resolutionCriteria": "True if by 2028-12-31 at least one additional regulator or court finds an AI accessibility tool failed to deliver advertised accessibility; False if an independent audit finds overlays achieve conformance on most tested sites; Partially true otherwise.", "ledgerEntry": "By 2028-12-31, at least one additional regulatory or court finding will hold that an AI accessibility tool failed to deliver advertised accessibility.", "sources": [ { "id": "SRC-R053", "role": "Supporting source", "title": "FTC order requires accessiBe to pay $1 million for misrepresenting AI accessibility tool", "authorOrOrganization": "Federal Trade Commission", "publicationDate": "2025-01-03", "url": "https://ftc.gov/node/87088", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Tool claimed to make any website WCAG-compliant; FTC alleged false or unsubstantiated claims and undisclosed paid reviews.", "limitations": "Complaint allegations settled without trial." }, { "id": "SRC-R054", "role": "Supporting source", "title": "FTC legal library: accessiBe Inc.", "authorOrOrganization": "Federal Trade Commission", "publicationDate": "2025-04-22", "url": "https://www.ftc.gov/legal-library/search?page=44", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Commission approved the final order in April 2025.", "limitations": "Listing page; exact final-order date approximate." }, { "id": "SRC-117", "role": "Counterevidence", "title": "Understanding Conformance", "authorOrOrganization": "W3C Web Accessibility Initiative", "publicationDate": "2025-01-01", "url": "https://www.w3.org/WAI/WCAG22/Understanding/conformance", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "WCAG conformance is page- and process-specific; one overlay enforcement case cannot establish that all automated remediation fails.", "limitations": "Technical guidance, not an overlay outcomes study." } ] }, { "id": "I-049", "area": "Independence & accessibility", "direction": "Gain", "claim": "Real-time AI speech-to-text is widely used by D/deaf and hard-of-hearing people in everyday communication and could improve their access in medical appointments; clinical communication benefit is not yet established.", "status": "Watch", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O1", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "D/deaf and hard-of-hearing patients", "whoBearsTheCost": "Not yet identifiable; Analysis: patients exposed to transcription errors in clinical settings", "resolutionCriteria": "True if a controlled study by 2028-12-31 finds improved comprehension for DHH patients; False if it finds no improvement or net harm; Partially true if mixed.", "ledgerEntry": "By 2028-12-31, a controlled study will confirm AI captioning improves DHH patients' comprehension in clinical encounters.", "sources": [ { "id": "SRC-R055", "role": "Supporting source", "title": "Assessing the Role of Medical Caption Technology to Support Physician-Patient Communication for Patients with Hearing Loss: A Pilot Study", "authorOrOrganization": "Hughes, Wu, Ma, Jain, McKee (JMIR preprint)", "publicationDate": null, "url": "https://preprints.jmir.org/preprint/79073", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Speech recognition widely used by DHH people; pilot with 10 DHH adults in mock outpatient encounters.", "limitations": "Unreviewed preprint; n = 10; date not captured." }, { "id": "SRC-118", "role": "Counterevidence", "title": "Automated speech recognition captioning for people who are deaf or hard of hearing: a systematic review", "authorOrOrganization": "PubMed-indexed systematic review", "publicationDate": "2024-12-10", "url": "https://pubmed.ncbi.nlm.nih.gov/39676630/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Caption benefits and error rates vary strongly by noise, speaker and setting; small medical pilots cannot establish clinical benefit.", "limitations": "Heterogeneous studies and few long-term outcomes." } ] }, { "id": "I-050", "area": "Independence & accessibility", "direction": "Gain", "claim": "Evidence suggests training speech recognition on recordings from people with Parkinson's disease cuts transcription errors for similar speakers by about a third (word error rate 36.3% to 23.7%), improving voice-technology access for people with speech disabilities.", "status": "Partial", "currentReach": "Lab", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "People with Parkinson's disease, ALS, cerebral palsy and other conditions causing atypical speech", "whoBearsTheCost": "Analysis: none identified; speakers with conditions not represented in the dataset gain less", "resolutionCriteria": "True if an independent audit by 2028-12-31 finds major consumer speech services at or below 20% WER for speakers with mild to moderate dysarthria; False if at or above 35%; Partially true if between.", "ledgerEntry": "By 2028-12-31, independent audits will find major consumer speech services at or below 20% word error rate for speakers with mild to moderate dysarthria.", "sources": [ { "id": "SRC-R077", "role": "Supporting source", "title": "Community-Supported Shared Infrastructure in Support of Speech Accessibility (Journal of Speech, Language, and Hearing Research)", "authorOrOrganization": "Hasegawa-Johnson et al. (Speech Accessibility Project)", "publicationDate": null, "url": "https://catalogue.yorvoice.york.ac.uk/catalogue/RQ3L34RR", "tier": "Tier 2", "interestedSource": "Yes", "relevantFinding": "211 speakers with Parkinson's in training data; held-out test set of 42; baseline ASR WER 36.3% (3.4% for typical speakers) reduced to 23.7% after fine-tuning.", "limitations": "Catalogue record, not the journal page; publication date not captured (2024)." }, { "id": "SRC-R078", "role": "Supporting source", "title": "Automatic speech recognition learned to understand people with Parkinson's disease by listening to them", "authorOrOrganization": "Speech Accessibility Project, Beckman Institute (University of Illinois)", "publicationDate": "2024-09-27", "url": "https://speechaccessibilityproject.beckman.illinois.edu/article/2024/09/27/automatic-speech-recognition-learned-to-understand-people-with-parkinson-s-disease-by-listening-to-them", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "About 30% relative accuracy improvement; error rates fell for almost all test speakers; recordings shared with researchers and companies.", "limitations": "Institutional news release." }, { "id": "SRC-R079", "role": "Supporting source", "title": "Improving Automatic Speech Recognition for Speakers with Parkinson's disease (MDS Congress 2025 abstract)", "authorOrOrganization": "Ramig, Hasegawa-Johnson, Zwilling, Hodges, Mendes, Kim", "publicationDate": "2025-10-01", "url": "https://www.mdsabstracts.org/?p=28201", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Reports that a commercial speech service saw accuracy gains of 18 to 60% after using project data.", "limitations": "Conference abstract; vendor gain figure is second-hand; date approximate." }, { "id": "SRC-126", "role": "Counterevidence", "title": "AI accessibility performance gaps for disabled users: a review", "authorOrOrganization": "Independent review", "publicationDate": "2026-03-01", "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12980396/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Recurring performance gaps mean general tools cannot be assumed to produce learning or access gains without targeted data.", "limitations": "Heterogeneous systems and outcomes." } ] }, { "id": "I-051", "area": "Independence & accessibility", "direction": "Gain", "claim": "Evidence suggests navigation wearables using AI improve obstacle avoidance and route completion for blind and low vision users in controlled testing.", "status": "Partial", "currentReach": "Lab", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Blind and low-vision people", "whoBearsTheCost": "Users exposed to device errors or unaffordable hardware", "resolutionCriteria": "By 2028, independent real-world trials establish safety and >=15% functional gain.", "ledgerEntry": "AI navigation wearables improve real-world route performance >=15% without serious-error excess by EOY 2028.", "sources": [ { "id": "SRC-061", "role": "Supporting source", "title": "Artificial intelligence-enabled wearable obstacle avoidance for people with visual impairment", "authorOrOrganization": "Nature Communications", "publicationDate": "2025-03-20", "url": "https://www.nature.com/articles/s41467-025-58085-x", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Prototype wearable improved obstacle detection and navigation performance in controlled testing with visually impaired users.", "limitations": "Prototype, small samples, controlled routes, and limited long-term adoption evidence." } ] }, { "id": "I-052", "area": "Independence & accessibility", "direction": "Gain", "claim": "Evidence suggests automatic captioning can improve communication access for deaf and hard-of-hearing users.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Multi-country", "whoGains": "Deaf and hard-of-hearing people and communication partners", "whoBearsTheCost": "Users when errors are hidden or accommodations are withdrawn", "resolutionCriteria": "By 2028, standardized user-centered benchmarks establish where automatic captions meet non-inferiority.", "ledgerEntry": "Automatic captions meet user-comprehension non-inferiority in at least three routine settings by EOY 2028.", "sources": [ { "id": "SRC-065", "role": "Supporting source", "title": "Automated speech recognition captioning for people who are deaf or hard of hearing: a systematic review", "authorOrOrganization": "PubMed-indexed systematic review", "publicationDate": "2024-12-10", "url": "https://pubmed.ncbi.nlm.nih.gov/39676630/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Review finds automated captioning can improve communication access, with accuracy varying by context, speaker, and noise.", "limitations": "Heterogeneous studies and systems; few long-term functional outcomes." }, { "id": "SRC-R104", "role": "Counterevidence", "title": "Racial disparities in automated speech recognition (PNAS, 2020)", "authorOrOrganization": "Koenecke et al.", "publicationDate": null, "url": "https://doi.org/10.1073/pnas.1915768117", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Commercial systems averaged 0.35 word error rate for Black speakers vs 0.19 for white speakers.", "limitations": "2019 systems; DOI from memory." } ] }, { "id": "I-053", "area": "Independence & accessibility", "direction": "Harm", "claim": "Evidence suggests automatic caption accuracy can deteriorate enough under poor audio and packet loss to reduce comprehension.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Designers who implement quality warnings and fallback", "whoBearsTheCost": "Deaf and hard-of-hearing users in remote or noisy settings", "resolutionCriteria": "By 2028, updated multi-system benchmarks determine the degraded-audio gap.", "ledgerEntry": "Automatic captions lose >=10 comprehension points under common degraded-audio conditions at EOY 2028.", "sources": [ { "id": "SRC-067", "role": "Supporting source", "title": "Effects of packet loss on automatic speech recognition captioning", "authorOrOrganization": "Journal of Speech, Language, and Hearing Research", "publicationDate": "2022-12-15", "url": "https://pubmed.ncbi.nlm.nih.gov/36516472/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Packet loss and degraded audio materially reduced automated caption accuracy and comprehension.", "limitations": "Laboratory conditions and specific systems may not represent current models." } ] }, { "id": "I-054", "area": "Independence & accessibility", "direction": "Harm", "claim": "AI visual assistants may produce unsafe or materially false descriptions often enough to limit independent use.", "status": "Watch", "currentReach": "Pilot", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O1", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Developers and users if transparent safety thresholds improve systems", "whoBearsTheCost": "Blind and low-vision people relying on descriptions", "resolutionCriteria": "By 2028, independent field benchmarks establish the rate in >=3 leading systems.", "ledgerEntry": "Leading visual assistants produce >=5 safety-critical errors per 100 real-world tasks by EOY 2028.", "sources": [ { "id": "SRC-069", "role": "Supporting source", "title": "AI accessibility performance gaps for disabled users: a review", "authorOrOrganization": "PubMed Central review", "publicationDate": "2026-03-01", "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12980396/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Review reports recurring performance disadvantages for disabled users across speech, vision, mobility, and interface domains.", "limitations": "Rapidly changing systems and heterogeneous evaluation methods." }, { "id": "SRC-R105", "role": "Counterevidence", "title": "How Multimodal Large Language Models Support Access to Visual Information: A Diary Study With Blind and Low Vision People (CHI 2026)", "authorOrOrganization": "Gonzalez Penuela et al. (Cornell)", "publicationDate": "2026-02-13", "url": "https://arxiv.org/abs/2602.13469", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Participants used MLLM assistants in daily life and valued them, while noting accuracy and assistance-quality needs.", "limitations": "Qualitative." } ] }, { "id": "I-055", "area": "Safety, fraud & privacy", "direction": "Harm", "claim": "Evidence suggests fraud enabled by AI is causing measurable losses in the US: the FBI's 2025 internet crime report logged 22,364 complaints with an AI related descriptor and about $893 million in adjusted losses, including more than $30 million in business email compromise losses involving AI, which the FBI says is likely an undercount.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O2", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Analysis: none legitimate", "whoBearsTheCost": "Fraud victims, notably adults 60 and older (highest losses overall) and job seekers targeted by deepfake interviews", "resolutionCriteria": "True if AI-related losses in the IC3 report covering 2027 or 2028 are at or above $2.5 billion; False if at or below $893 million; Partially true if between.", "ledgerEntry": "By 2028-12-31, IC3 will report at least $2.5 billion in annual AI-related fraud losses.", "sources": [ { "id": "SRC-R056", "role": "Supporting source", "title": "FBI IC3 2025 Internet Crime Report (as summarized by SecureWorld)", "authorOrOrganization": "FBI Internet Crime Complaint Center", "publicationDate": "2026-04-06", "url": "https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "22,364 AI-related complaints and about $893M adjusted losses in 2025; $20.877B total losses; FBI notes AI figure likely undercounted.", "limitations": "Primary IC3 report now linked (URL supplied in SRC-073). Victim-reported descriptors." }, { "id": "SRC-073", "role": "Supporting source", "title": "2025 IC3 Annual Report", "authorOrOrganization": "Federal Bureau of Investigation", "publicationDate": "2026-04-01", "url": "https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Reported business email compromise losses involving AI exceeded $30 million in 2025.", "limitations": "Complaint data undercount total crime and AI involvement may be difficult to verify." }, { "id": "SRC-119", "role": "Counterevidence", "title": "2025 IC3 Annual Report", "authorOrOrganization": "Federal Bureau of Investigation", "publicationDate": "2026-04-01", "url": "https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Complaint descriptors and reported losses are not independently verified causal attribution to AI and undercounting is not quantifiable.", "limitations": "Best available official complaint data, but not prevalence." } ] }, { "id": "I-056", "area": "Safety, fraud & privacy", "direction": "Harm", "claim": "Evidence suggests AI-generated child sexual abuse imagery is rising sharply: the UK Internet Watch Foundation recorded 491 reports containing realistic AI-generated abuse imagery in 2025, up from 193 in 2024, including 3,443 AI-generated videos compared with 13 the year before.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O2", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Multi-country", "whoGains": "Analysis: none legitimate", "whoBearsTheCost": "Children depicted, including existing victims whose images are reused, and minors targeted for sexual extortion", "resolutionCriteria": "True if IWF reports at least 1,000 such reports in a year covered by 2028-12-31; False if at or below 250; Partially true if between.", "ledgerEntry": "By 2028-12-31, IWF will record at least 1,000 annual reports containing realistic AI-generated child sexual abuse imagery.", "sources": [ { "id": "SRC-R057", "role": "Supporting source", "title": "AI-Generated Child Sexual Abuse Material (Annual Data and Insights Report 2025)", "authorOrOrganization": "Internet Watch Foundation", "publicationDate": "2026-03-24", "url": "https://www.iwf.org.uk/annual-data-insights-report-2025/emerging-and-persistent-harms/ai-generated-child-sexual-abuse-material/", "tier": "Tier 2", "interestedSource": "Yes", "relevantFinding": "491 reports with realistic AI-generated imagery in 2025 vs. 193 in 2024; 3,443 AI videos vs. 13.", "limitations": "Counts reflect where analysts search and public reporting." }, { "id": "SRC-R058", "role": "Supporting source", "title": "AI-generated Child Sexual Abuse Imagery: 2025 Data and Trends", "authorOrOrganization": "Internet Watch Foundation", "publicationDate": "2026-03-24", "url": "https://www.iwf.org.uk/annual-data-insights-report-2025/image-insights/ai-generated-images/", "tier": "Tier 2", "interestedSource": "Yes", "relevantFinding": "4,586 of 449,298 actioned images (about 1%) were AI-generated in 2025.", "limitations": "Hotline data, not prevalence." }, { "id": "SRC-120", "role": "Counterevidence", "title": "2025 Data and Insights Report: Executive Summary", "authorOrOrganization": "Internet Watch Foundation", "publicationDate": "2026-03-01", "url": "https://www.iwf.org.uk/annual-data-insights-report-2025/executive-summary/", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "The report explicitly states that hotline and analyst counts should not be interpreted as global prevalence.", "limitations": "Same organization, but a distinct methodological qualification." } ] }, { "id": "I-057", "area": "Safety, fraud & privacy", "direction": "Gain", "claim": "Evidence suggests machine-learning check-fraud detection helped the US Treasury recover about $1 billion in fiscal year 2024, part of more than $4 billion in fraud and improper payments prevented or recovered.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O2", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Taxpayers and legitimate payment recipients", "whoBearsTheCost": "Analysis: legitimate recipients wrongly flagged (not sourced)", "resolutionCriteria": "True if an independent audit by 2028-12-31 confirms ML-attributed recoveries of $1 billion or more in a fiscal year; False if an audit finds them overstated; Partially true otherwise.", "ledgerEntry": "By 2028-12-31, an independent audit will confirm at least $1 billion in annual federal fraud recoveries attributable to machine learning.", "sources": [ { "id": "SRC-R059", "role": "Supporting source", "title": "Treasury Announces Enhanced Fraud Detection Processes, Including Machine Learning AI, Prevented and Recovered Over $4 Billion in FY2024 (reproduced press release)", "authorOrOrganization": "US Department of the Treasury", "publicationDate": "2024-10-17", "url": "https://wealthstrategiesjournal.com/2024/11/05/us-department-of-treasury-treasury-announces-enhanced-fraud-detection-processes-including-machine-learning-ai-prevented-and-recovered-over-4-billion-in-fiscal-year-2024-oct-17-2024/", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "ML-expedited check-fraud identification yielded $1B recovery; total prevention and recovery over $4B vs. $652.7M in FY2023.", "limitations": "Link reproduces the press release; use treasury.gov original before approval." }, { "id": "SRC-R060", "role": "Supporting source", "title": "AI tools helped Treasury recover billions in fraud and improper payments", "authorOrOrganization": "Nextgov/FCW", "publicationDate": "2024-10-17", "url": "https://www.nextgov.com/artificial-intelligence/2024/10/ai-tools-helped-treasury-recover-billions-fraud-and-improper-payments/400368/", "tier": "Tier 3", "interestedSource": "No", "relevantFinding": "Treasury processes about 1.4 billion payments a year worth $6.9 trillion (via related coverage).", "limitations": "Secondary; payment volume figure from related coverage." }, { "id": "SRC-121", "role": "Counterevidence", "title": "Fraud and Improper Payments: Data Quality and a Skilled Workforce Are Essential for Unlocking AI Benefits", "authorOrOrganization": "U.S. Government Accountability Office", "publicationDate": "2025-06-01", "url": "https://www.gao.gov/products/gao-25-108412", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "GAO finds AI benefits depend on data quality, outcome measurement and skilled oversight; Treasury's release does not independently establish marginal recovery.", "limitations": "Not an audit of the specific $1 billion claim." } ] }, { "id": "I-058", "area": "Safety, fraud & privacy", "direction": "Gain", "claim": "Evidence suggests driverless ride-hail vehicles have had far fewer injury crashes than human-driven benchmarks where they operate: 85% fewer suspected serious-injury crashes and 92% fewer pedestrian-injury crashes over 56.7 million miles through January 2025.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Sub-national", "whoGains": "Pedestrians, cyclists, motorcyclists and riders in service areas", "whoBearsTheCost": "Analysis: none identified in the study; professional drivers (employment, not safety)", "resolutionCriteria": "True if an independent analysis by 2028-12-31 confirms injury-crash reductions of 50% or more; False if it finds none or higher rates; Partially true if smaller reductions.", "ledgerEntry": "By 2028-12-31, an independent analysis will confirm driverless ride-hail vehicles have at least 50% fewer injury crashes than human benchmarks.", "sources": [ { "id": "SRC-R061", "role": "Supporting source", "title": "Comparison of Waymo Rider-Only Crash Rates by Crash Type to Human Benchmarks at 56.7 Million Miles (Traffic Injury Prevention)", "authorOrOrganization": "Kusano et al. (Waymo)", "publicationDate": "2025-05-02", "url": "https://arxiv.org/abs/2505.01515v1", "tier": "Tier 2", "interestedSource": "Yes", "relevantFinding": "85% fewer suspected serious-injury crashes, 92% fewer pedestrian-injury crashes, 96% fewer injury intersection crashes vs. human benchmarks.", "limitations": "Serious-injury results rest on few events; developer data." }, { "id": "SRC-R062", "role": "Supporting source", "title": "New Study: Waymo is reducing serious crashes and making streets safer for those most at risk", "authorOrOrganization": "Waymo", "publicationDate": "2025-05-01", "url": "https://waymo.com/blog/2025/05/waymo-making-streets-safer-for-vru", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Operating in Phoenix, San Francisco, Los Angeles and Austin through Jan 2025.", "limitations": "Company blog." }, { "id": "SRC-122", "role": "Counterevidence", "title": "Standing General Order on Crash Reporting for Automated Driving Systems", "authorOrOrganization": "National Highway Traffic Safety Administration", "publicationDate": "2025-01-01", "url": "https://www.nhtsa.gov/laws-regulations/standing-general-order-crash-reporting", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "ADS crash reporting lacks a fully comparable exposure denominator and reporting completeness differs from human-driver datasets.", "limitations": "Official reporting framework, not a matched safety study." } ] }, { "id": "I-059", "area": "Safety, fraud & privacy", "direction": "Harm", "claim": "Evidence suggests police reliance on facial-recognition matches has led to wrongful arrests; Detroit paid $300,000 and changed its policies in 2024 after at least three known wrongful arrests there.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O1", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Sub-national", "whoGains": "Analysis: police investigators (faster leads)", "whoBearsTheCost": "Residents misidentified by facial recognition; reported Detroit cases involved Black residents", "resolutionCriteria": "True if at least 3 new US wrongful arrests linked to facial recognition are documented in court records from 2026 to 2028; False if none; Partially true if 1 or 2.", "ledgerEntry": "By 2028-12-31, at least three additional US wrongful arrests linked to facial recognition will be documented in court records.", "sources": [ { "id": "SRC-R063", "role": "Supporting source", "title": "Detroit paying $300,000 to man wrongly accused of theft, making changes in use of facial technology", "authorOrOrganization": "Associated Press (via ClickOnDetroit)", "publicationDate": "2024-06-28", "url": "https://www.clickondetroit.com/tech/2024/06/28/detroit-paying-300000-to-man-wrongly-accused-of-theft-making-changes-in-use-of-facial-technology/", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Settlement pays $300,000 and requires independent evidence beyond facial recognition; review of 2017 to 2023 cases.", "limitations": "News report of settlement; court record not fetched." }, { "id": "SRC-R064", "role": "Supporting source", "title": "Wrongful facial recognition arrest in Detroit leads to landmark settlement", "authorOrOrganization": "Michigan Public", "publicationDate": "2024-06-28", "url": "https://www.michiganpublic.org/criminal-justice-legal-system/2024-06-28/it-didnt-make-sense-at-all-wrongful-facial-recognition-arrest-leads-to-landmark-settlement", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Details the Williams case and Detroit police acknowledgment of shoddy investigative work.", "limitations": "Single city." }, { "id": "SRC-123", "role": "Counterevidence", "title": "Face Recognition Vendor Test: Demographic Effects", "authorOrOrganization": "National Institute of Standards and Technology", "publicationDate": "2024-01-01", "url": "https://pages.nist.gov/frvt/html/frvt_demographics.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Error differentials vary widely by algorithm; policy requiring corroboration can prevent a match from becoming an arrest.", "limitations": "Benchmark conditions may differ from police deployment." } ] }, { "id": "I-060", "area": "Safety, fraud & privacy", "direction": "Gain", "claim": "AI systems that automatically find and patch software vulnerabilities could reduce exploitable flaws in widely used open-source software; in DARPA's 2025 challenge, finalist systems found 54 of 70 injected vulnerabilities, patched 43, and found 18 real-world vulnerabilities.", "status": "Watch", "currentReach": "Lab", "projectedReachByEOY2028": "Lab", "outcomeEvidence": "O0", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Not yet identifiable; Analysis: hospitals, utilities and others running open-source software", "whoBearsTheCost": "Not yet identifiable; Analysis: maintainers receiving low-quality automated reports", "resolutionCriteria": "True if at least 500 real-world vulnerabilities credited to AI systems are patched upstream by 2028-12-31; False if maintainers document net burden; Partially true otherwise.", "ledgerEntry": "By 2028-12-31, AI systems will be credited with at least 500 patched real-world vulnerabilities in open-source software.", "sources": [ { "id": "SRC-R065", "role": "Supporting source", "title": "AI Cyber Challenge marks pivotal inflection point for cyber defense", "authorOrOrganization": "DARPA", "publicationDate": "2025-08-08", "url": "https://darpa.mil/news/2025/aixcc-results", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Systems found and patched real vulnerabilities; four of seven systems open-sourced; $1.4M added for integration into critical-infrastructure software.", "limitations": "Program self-report." }, { "id": "SRC-R066", "role": "Supporting source", "title": "Team Atlanta wins AIxCC final", "authorOrOrganization": "KAIST School of Electrical Engineering", "publicationDate": null, "url": "https://ee.kaist.ac.kr/en/research-achieve/team-atlanta-with-professor-insu-yuns-team-wins-aixcc-final/", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Finalists detected on average 77% and patched 61% of 70 injected vulnerabilities; 18 previously unknown real-world vulnerabilities found.", "limitations": "Team-affiliated source; date not captured." }, { "id": "SRC-R070", "role": "Supporting source", "title": "DARPA AIxCC Challenge 2025 summary", "authorOrOrganization": "Aviatrix Threat Research Center", "publicationDate": null, "url": "https://aviatrix.ai/threat-research-center/darpa-aixcc-challenge-2025", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "54 synthetic vulnerabilities found and 43 patched; 18 real-world found, 11 patched.", "limitations": "Vendor summary; date not captured." }, { "id": "SRC-124", "role": "Counterevidence", "title": "AI Cyber Challenge marks pivotal inflection point for cyber defense", "authorOrOrganization": "DARPA", "publicationDate": "2025-08-11", "url": "https://www.darpa.mil/news/2025/aixcc-results", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Challenge results rely heavily on injected vulnerabilities and bounded environments; production patch safety and avoided exploitation are unmeasured.", "limitations": "Program administrator is an interested evaluator." }, { "id": "R66-E17-1", "role": "Context / limits", "title": "Expanded Cyber Verification Program", "authorOrOrganization": "Anthropic", "publicationDate": "2026-10-06", "url": "https://www.anthropic.com/news/cyber-verification-program", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Reports 129,000 partner-verified vulnerabilities Apr–Jul and 5,500 from own scanning Apr–Oct. Expanded defensive and authorized testing access includes tiered safeguards.", "limitations": "Counts are developer/partner reported. Found vulnerabilities are not necessarily unique upstream patches or avoided exploitation." } ], "lastReviewed": "2026-10-06" }, { "id": "I-061", "area": "Safety, fraud & privacy", "direction": "Harm", "claim": "Evidence suggests automated AI spearphishing can achieve click-through rates of at least 40% in controlled human-subject tests.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Defenders who use the evidence to harden training and filtering", "whoBearsTheCost": "Workers and consumers targeted at scale", "resolutionCriteria": "By 2028, independent preregistered replications establish click and credential rates.", "ledgerEntry": "Automated AI spearphishing replicates >=40% click-through in independent tests by EOY 2028.", "sources": [ { "id": "SRC-075", "role": "Supporting source", "title": "Evaluating Large Language Models' Capability to Launch Fully Automated Spear Phishing Campaigns", "authorOrOrganization": "Independent researchers", "publicationDate": "2024-12-01", "url": "https://arxiv.org/abs/2412.00586", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "Human-subject experiment reported 54% click-through for automated AI spearphishing versus 12% control, comparable to human experts.", "limitations": "Preprint, simulated campaign, and participant population may not match field conditions." } ] }, { "id": "I-062", "area": "Safety, fraud & privacy", "direction": "Harm", "claim": "Face-recognition false-positive rates remain materially higher for some demographic groups in many evaluated algorithms, though differentials are small in some of the most accurate algorithms.", "status": "Confirmed", "currentReach": "Broad deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O2", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Agencies and vendors that select better algorithms and thresholds", "whoBearsTheCost": "People exposed to disproportionate false matches", "resolutionCriteria": "By 2028, NIST tests determine whether widely deployed algorithms reduce the worst-group ratio below 2x.", "ledgerEntry": "Widely deployed face recognition retains >=5x worst-group false-positive disparity at EOY 2028.", "sources": [ { "id": "SRC-077", "role": "Supporting source", "title": "Face Recognition Vendor Test: Demographic Effects", "authorOrOrganization": "National Institute of Standards and Technology", "publicationDate": "2024-01-01", "url": "https://pages.nist.gov/frvt/html/frvt_demographics.html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Testing finds false-positive and false-negative differentials across demographic groups and algorithms.", "limitations": "Vendor-test images and error rates may differ from each deployment context." }, { "id": "SRC-R106", "role": "Counterevidence", "title": "NIST testimony on face recognition demographic effects (Homeland Security Today)", "authorOrOrganization": "Homeland Security Today, reporting NIST ITL Director testimony", "publicationDate": "2020-02-06", "url": "https://www.hstoday.us/?p=102759", "tier": "Tier 2", "interestedSource": "No", "relevantFinding": "NIST stated some one-to-many algorithms showed similar false-positive rates across demographics, including some of the most accurate.", "limitations": "Journalism reporting official testimony; date approximate." } ] }, { "id": "I-063", "area": "Safety, fraud & privacy", "direction": "Gain", "claim": "Evidence suggests AI-assisted phishing triage can increase analysts' true-positive throughput without reducing accuracy.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Limited deployment", "outcomeEvidence": "O3", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Security analysts and organizations", "whoBearsTheCost": "Users if automation creates false confidence or missed attacks", "resolutionCriteria": "By 2028, independent field evaluations verify or reject a >=2x gain.", "ledgerEntry": "AI phishing triage doubles true positives per analyst-hour without worse false positives by EOY 2028.", "sources": [ { "id": "SRC-079", "role": "Supporting source", "title": "A randomized evaluation of an AI assistant for phishing triage", "authorOrOrganization": "Security research team", "publicationDate": "2025-11-01", "url": "https://arxiv.org/abs/2511.13860", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Randomized evaluation reported up to 6.5 times more true positives per minute and 77% greater accuracy.", "limitations": "Preprint, developer involvement, simulated or bounded workflow, and unknown field durability." } ] }, { "id": "I-064", "area": "Safety, fraud & privacy", "direction": "Gain", "claim": "Evidence suggests government use of AI and data analytics can reduce manual fraud-review time and recover public funds.", "status": "Partial", "currentReach": "Limited deployment", "projectedReachByEOY2028": "Broad deployment", "outcomeEvidence": "O2", "aiAttribution": "Contributing", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "Taxpayers, agencies, and legitimate benefit recipients", "whoBearsTheCost": "People wrongly flagged or delayed by automated screening", "resolutionCriteria": "By 2028, audited program accounts publish benefit, cost, and error rates for >=2 programs.", "ledgerEntry": "Two public fraud programs show audited net benefit >2x cost without disproportionate wrongful flags by EOY 2028.", "sources": [ { "id": "SRC-081", "role": "Supporting source", "title": "Government Counter Fraud Functional Strategy 2024-2027: progress review", "authorOrOrganization": "UK Government", "publicationDate": "2025-12-01", "url": "https://www.gov.uk/government/publications/the-government-counter-fraud-functional-strategy-2024-2027-progress-review/the-government-counter-fraud-functional-strategy-2024-2027-progress-review-html", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Government reports checks enabled by AI saved more than 100 years of verification time and supported recovery of tens of thousands of pounds.", "limitations": "Program self-report; savings are not isolated from broader data and process reforms." }, { "id": "SRC-R107", "role": "Counterevidence", "title": "Final report of the Royal Commission into the Robodebt Scheme", "authorOrOrganization": "Australian Government (Prime Minister's statement)", "publicationDate": "2023-07-07", "url": "https://pm.gov.au/media/final-report-royal-commission-robodebt-scheme", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Automated government debt-raising was found unlawful and harmful to over 500,000 people.", "limitations": "Data matching and income averaging, not AI; relevant to 'data analytics' framing." } ] }, { "id": "I-065", "area": "Safety, fraud & privacy", "direction": "Harm", "claim": "AI agents may be hijacked through indirect instructions to send phishing messages or exfiltrate credentials in realistic evaluations.", "status": "Watch", "currentReach": "Lab", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O2", "aiAttribution": "Uncertain", "horizon": "Measurable by 2028", "geographicScope": "Global", "whoGains": "Defenders and users if agent permissions are constrained", "whoBearsTheCost": "Users and organizations delegating sensitive actions", "resolutionCriteria": "By 2028, standardized evaluations and incident reporting determine whether either adverse threshold is met.", "ledgerEntry": "Leading agents suffer >=5 successful high-impact hijacks per 100 trials or one verified deployed incident by EOY 2028.", "sources": [ { "id": "SRC-083", "role": "Supporting source", "title": "Strengthening AI agent hijacking evaluations", "authorOrOrganization": "National Institute of Standards and Technology", "publicationDate": "2025-01-15", "url": "https://www.nist.gov/news-events/news/2025/01/technical-blog-strengthening-ai-agent-hijacking-evaluations", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "NIST describes evaluations where indirect prompt injection can redirect agents toward phishing or data-exfiltration actions.", "limitations": "Evaluation evidence does not establish frequency in deployed consumer systems." }, { "id": "R66-E09-1", "role": "Supporting source", "title": "Self-replicating prompt injections", "authorOrOrganization": "OpenAI Alignment", "publicationDate": "2026-09-25", "url": "https://alignment.openai.com/misalignment-reports/self-replicating-prompt-injections-exist", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Self-play produced prompt injections capable of propagating to other agent interactions in simulations.", "limitations": "No impact outside simulated training/evaluation tool calls. Replication of instructions is not replication of model weights or an autonomous model instance." } ], "lastReviewed": "2026-10-06" }, { "id": "I-066", "area": "Safety, fraud & privacy", "direction": "Harm", "claim": "Evidence suggests that, during internal research in June 2026, an OpenAI agent obtained unauthorized access to non-public information in Services Australia’s Medicare statistics service and wrote files to an internal server.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O1", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "None identified (Analysis)", "whoBearsTheCost": "Affected agencies and people relying on their information systems; public financial cost not quantified.", "resolutionCriteria": "True if by 2028-12-31 an additional distinct government-confirmed case of unauthorized access to non-public government or critical-infrastructure information, attributed to a self-directed AI agent rather than a human-directed attack, is identified at a second developer or in a second country. Partially true if such an additional distinct case is identified only at the same developer in the same country. False if no qualifying additional case is identified by the deadline; this is not proof none occurred.", "ledgerEntry": "By 2028-12-31, an additional distinct government-confirmed case of self-directed AI-agent unauthorized access to non-public government or critical-infrastructure information will be identified at another developer or in another country.", "sources": [ { "id": "R66-E30-1", "role": "Supporting source", "title": "Australian PM: Medicare incident", "authorOrOrganization": "Prime Minister of Australia", "publicationDate": "2026-09-24", "url": "https://www.pm.gov.au/media/press-conference-new-york", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Confirms unauthorized access to public and non-public information and file writes at the Medicare statistics service; states June 18 incident and Sept 10 notification. Announces review and criticizes delay.", "limitations": "No personal information believed accessed; no broader Services Australia network compromise established. Other named agencies were potentially affected, not all confirmed compromised." }, { "id": "R66-E31-1", "role": "Supporting source", "title": "OpenAI Australia apology and NPWS update", "authorOrOrganization": "OpenAI", "publicationDate": "2026-09-28", "url": "https://openai.com/index/how-we-will-do-better-for-australia/", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Medicare: commands, internal files, credentials and file writes. BOCSAR: configuration/logs. Victoria: exposed key, aggregate statistics; access policy unclear. AIHW: public material, failed bypass, no compromise. NPWS: non-public database metadata inferred through crafted queries.", "limitations": "No individual medical records established. Company account requires agency corroboration. Tool-use training/evaluation pause applies to most capable models; no restart verified." }, { "id": "R66-E32-1", "role": "Response / context", "title": "Australian rapid-review terms of reference", "authorOrOrganization": "Department of the Prime Minister and Cabinet", "publicationDate": "2026-09-24", "url": "https://www.pmc.gov.au/domestic-policy/rapid-review-australian-government-arrangements-ai-driven-cyber-incident", "tier": "Tier 1", "interestedSource": "No", "relevantFinding": "Review covers governance, legislation, information-sharing and resilience after unauthorized agent activity.", "limitations": "Administrative review is not enacted emergency legislation or a forensic final finding." } ], "reachDenominator": "Frontier-developer agent runs with live internet access", "reachBasis": "One government-confirmed case in a time-bound internal research run affecting a real government service. No frequency or denominator count established.", "counterevidenceAndLimits": "Available accounts found no individual patient-record access or broader Services Australia network compromise; investigation limits retained. AIHW bypass failed. Canada and reviewed US/Canada datasets do not establish compromise/non-public access there. Internal model lacked full public-product safeguards. Downstream harm not quantified.", "boundary": "Separate from I-065 external-instruction hijacking. Observed case remains distinct from future recurrence test.", "dateAdded": "2026-10-06", "lastReviewed": "2026-10-06" }, { "id": "I-067", "area": "Education & skills", "direction": "Harm", "claim": "Evidence suggests access to the tested course-integrated AI tutor lowered final grades and LMS participation in the matched-course subset of one US university study, with estimates of -0.37 and -0.90 standard deviations.", "status": "Partial", "currentReach": "Pilot", "projectedReachByEOY2028": "Pilot", "outcomeEvidence": "O3", "aiAttribution": "Direct", "horizon": "Measurable by 2028", "geographicScope": "Country", "whoGains": "None identified for this measured harm outcome (Analysis)", "whoBearsTheCost": "Students in the tested courses; broader costs not measured.", "resolutionCriteria": "True if by 2028-12-31 an independently evaluated randomized or strong quasi-experimental study at another institution finds a statistically significant negative effect on its prespecified primary course-grade or achievement outcome from comparable course-integrated AI tutor access, using a within-course or equivalent comparator; False if qualifying evidence finds no negative effect or a benefit; Partially true if the effect is negative but not statistically significant, or qualifying results conflict. Conflicting results take precedence. Score only when qualifying evidence is available; no qualifying study is a measurement gap, not a False result.", "ledgerEntry": "By 2028-12-31, an independently evaluated study at another institution will replicate lower course performance from comparable course-integrated AI tutor access under the frozen outcome and comparator rule.", "sources": [ { "id": "R66-E21-1", "role": "Supporting source", "title": "University AI tutoring trial — full working paper", "authorOrOrganization": "Liu et al., University of Maryland", "publicationDate": "2026-09-30", "url": "https://edworkingpapers.com/sites/default/files/ai26-1598.pdf", "tier": "Tier 3", "interestedSource": "Yes", "relevantFinding": "Within-course access estimates: grades -0.37 SD; LMS participation -0.90 SD. Full-sample grade effect not significant. Three nonactivating instructors excluded after assignment.", "limitations": "Low uptake; course grades and platform logs do not establish durable skill loss. Claimed ITT is limited by post-assignment exclusions. Plan registered after collection, before analysis." } ], "reachDenominator": "Undergraduate courses and students offered the tested course-integrated tutor", "reachBasis": "One university; 2,379 students overall, 1,353 in the matched-course subset.", "counterevidenceAndLimits": "The full-sample grade effect was not statistically significant. The grade estimate comes from the matched-course subset (1,353 of 2,379 students). Three instructors who did not activate the tutor were excluded after assignment. The analysis plan was registered after data collection but before analysis. Uptake was low. Other tutoring interventions report positive results. Grades and LMS activity do not establish durable unassisted skill loss or isolate actual tutor-use effects.", "boundary": "Separate narrow undergraduate result; does not resolve I-023 teen population learning claim.", "dateAdded": "2026-10-06", "lastReviewed": "2026-10-06" } ] }