Scores by kind of claim
Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.
-
Factual claims
Statements about the present or the past, checked against the record: did they hold up?
79 72 85 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
80 72 88 -
Promises
Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?
Insufficient record: 1 promise resolved, 5 needed for a score.
All types combined: 79 (likely 74 to 84, high confidence), from 102 claims: 64 factual claims, 38 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Gary Marcus's claims
“AI Safety Institutes will also offer guidance, but have little legal authority to stop truly dangerous models should they arise.”
Gary Marcus, Substack author 1 January 2025 25 AI Predictions for 2025, from Marcus on AI (Marcus on AI)
AI Safety Institutes will offer guidance in 2025 but have little legal authority to stop truly dangerous models.
Resolved true The record shows it happened or was so.
The second judge would not resolve this claim; the verdict shown is the first judge's.
What would count as true or false
Judge as of 31 Dec 2025. The claim has two parts: (a) AI Safety Institutes (US AISI/CAISI, UK AISI/AI Security Institute, and similar national bodies) offered guidance (published evaluations, guidance documents, voluntary frameworks, advice) during 2025; (b) through 2025 they had little legal authority to stop dangerous models, meaning no statutory power to block, recall or halt the deployment of an AI model. True: both hold, i.e. the institutes issued guidance in 2025 and none of the major ones (US, UK, plus most others) gained binding legal power to prevent a model's release or deployment by end-2025. Mostly true: both hold in substance, but one notable institute gained a limited binding power that it did not use or that covered only a narrow case, or guidance output was thin while the authority part plainly held. Mixed: one part clearly holds and the other clearly fails, e.g. institutes were abolished or stopped issuing guidance, while the lack of authority held. Mostly false: one or more major institutes (US or UK) gained real legal authority to block or halt dangerous models by end-2025, though some others did not. False: the major institutes held clear statutory power to stop model deployment during 2025 and also did not offer guidance. Powers held by other regulators (e.g. the EU AI Office under the AI Act) count only if the body is itself an AI Safety Institute or the claim's equivalent.
Evidence
- AI Safety Institutes Around the World AI Understanding (501(c)(3) nonprofit) · 2 October 2026
“A common misconception is that these institutes can block model releases. The UK and US bodies have no general statutory power to compel access or stop deployment.”
- UK AI Security Institute 2026: What Actually Governs Your Data Vucense · 20 February 2026
“It is not a licensing body. It has no statutory power to approve a deployment, block a release, or issue binding rulings to developers.”
- America's AI Action Plan The White House · July 2025
“Led by NIST at DOC, including CAISI, partner with the AI and cybersecurity industries to ensure”
- Inaugural report pioneered by AI Security Institute gives clearest picture yet of capabilities of most advanced AI GOV.UK (UK Government) · 18 December 2025
“The report is not intended to make policy recommendations, but gives decision makers in the UK and across the world the clearest data yet of precisely what the most advanced AI systems can actually do.”
Reasoning
On the guidance part, the UK AI Security Institute published an inaugural report on advanced AI capabilities in December 2025, giving decision makers evaluation data (E4). The 2025 US AI Action Plan also gives CAISI a role working with industry (E3). On the authority part, independent sources from 2026 say the UK and US institutes have no general statutory power to compel access or stop deployment (E1), and that the UK institute cannot approve deployments, block releases or issue binding rulings (E2). Both parts of the prediction held through 2025.
Cite this claim
Gary Marcus, “AI Safety Institutes will offer guidance in 2025 but have little legal authority to stop truly dangerous models.” (said 1 January 2025). Verdict: resolved true. Reliability API, claim c_1a49cb565e4e, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_1a49cb565e4e
Claims without a verdict
Is a verdict here wrong, or do you speak for Gary Marcus? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.