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?
77 71 83 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
58 38 77
All types combined: 75 (likely 70 to 81, high confidence), from 85 claims: 79 factual claims, 6 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Bill Gates's claims
10 claims
“I have to say I’m kind of shocked that the exact criteria that we review these models with, and the actions we take to minimize the harms, are really completely missing.”
Bill Gates, interviewed by CNN's Anderson Cooper 26 August 2026 Bill Gates says there needs to be limits on AI (CNN)
The exact criteria for reviewing AI models and the actions to minimize their harms are completely missing.
Mostly false False in substance, right in a detail.
What would count as true or false
Judge as of 2026-08-26 and in the context the speaker refers to (the identified models or developers, or the regulatory or disclosure setting). True if the evidence shows that no exact review criteria for evaluating these AI models, and no documented actions to minimize their harms, exist or have been disclosed in that context. Mostly true if such criteria and actions are almost entirely absent, with only minor or vague exceptions. Mixed if criteria or harm-mitigation actions exist for some of the models or developers in question but not others, or one of the two exists and the other does not. Mostly false if such criteria and actions are published, e.g. safety frameworks, model or system cards, evaluation reports, or regulatory requirements, but are incomplete or not exact, so 'completely missing' overstates the gap. False if detailed, published review criteria and harm-mitigation actions plainly exist for the models in question.
Evidence
- Guidelines for providers of general-purpose AI models European Commission (Digital Strategy) · 28 April 2026
“From 2 August 2026, the Commission’s enforcement powers enter into application. The Commission will enforce compliance with the obligations for providers of GPAI models, including with fines.”
- Guidelines for providers of general-purpose AI models European Commission (Digital Strategy) · 28 April 2026
“Safety and Security Framework and Model Report pursuant to Measures 1.4 and 7.7 of the Safety and Security Chapter of the GPAI Code of Practice”
- Guidelines for providers of general-purpose AI models European Commission (Digital Strategy) · 28 April 2026
“From 2 August 2025, the obligations for providers of GPAI models entered into application.”
- Guidelines for providers of general-purpose AI models European Commission (Digital Strategy) · 28 April 2026
“Thus, the guidelines complement the General-Purpose AI Code of Practice, a voluntary, practical tool that explains how providers of GPAI models can fulfil their obligations under the AI Act.”
- International AI Safety Report 2026 International AI Safety Report (Chair: Prof. Yoshua Bengio) · February 2026
“No established mechanism currently exists for validating risk acceptance decisions made by developers prior to release (1005)�”
- International AI Safety Report 2026 International AI Safety Report (Chair: Prof. Yoshua Bengio) · February 2026
“Recent frameworks and company policies have begun to formalise ‘risk acceptance’ criteria (965, 1040*)� However, setting appropriate thresholds remains challenging”
- International AI Safety Report 2026 International AI Safety Report (Chair: Prof. Yoshua Bengio) · February 2026
“implementing controls and countermeasures to reduce identified risks. Examples are access controls (991*), continuous monitoring (986), and if-then commitments (700)�”
- International AI Safety Report 2026 International AI Safety Report (Chair: Prof. Yoshua Bengio) · February 2026
“AI developers have information about their products that remains largely proprietary, and commercial considerations often make it difficult for them to share information about their development processes and risk assessments”
Reasoning
The evidence shows that review criteria and harm-mitigation measures for AI models exist in some form, including EU obligations for providers of general-purpose AI models and an accompanying Code of Practice [E3][E4], and frameworks and company policies that have begun to formalise risk acceptance criteria [E6], though much developer information remains proprietary [E8]. The EU's obligations for providers of general-purpose AI models took effect in August 2025, with enforcement powers from 2 August 2026. Providers are expected to submit a Safety and Security Framework and a Model Report under the GPAI Code of Practice (E1–E4). The International AI Safety Report 2026 says frameworks and company policies have begun to formalise risk-acceptance criteria and describes documented controls such as access controls and continuous monitoring (E6, E7). The same report notes real gaps: there is no established mechanism to validate developers' pre-release risk decisions, and much risk-assessment information stays proprietary (E5, E8). So the criteria and actions are incomplete rather than "completely missing", which fits mostly false.
Cite this claim
Bill Gates, “The exact criteria for reviewing AI models and the actions to minimize their harms are completely missing.” (said 26 August 2026). Verdict: mostly false. Reliability API, claim c_08d92eb05a57, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_08d92eb05a57
Claims without a verdict
Is a verdict here wrong, or do you speak for Bill Gates? 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.