100+ AI experts set minimum requirements for independent model evaluations
CFGeek · x · 2026-09-19
More than 100 leading AI experts endorsed a set of minimum requirements in response to AI companies' recent calls to embed external evaluators. The evaluators must be genuinely independent, transparent, and span a range of expertise; they need employee-level access and protection from retaliation for findings that make companies look bad. The experts welcome the labs' invitation of independent oversight but stress that what they do next matters — labs must be accountable for meeting these requirements so the public can trust the process.
Related event: 100+ AI experts set minimum bar for embedded third-party AI evaluations(8 posts)→
More from AGI Musings
- Gary Marcus: The near-term AI risk is agentic AI hacking the internet at scale, not rogue superintelligence — asusarla · 2026-09-19
- Researcher mocks time series benchmark culture: bechmaxxed models rarely match your domain — RexDouglass · 2026-09-19
- shlevy wavers: it's getting hard to defend that Dario truly means his AI safety warnings — NathanpmYoung · 2026-09-19
- Liam Fedus Amplifies: AI for Science Must Run Real-World Experiments — LiamFedus · 2026-09-19
- Mathematician's talk abstract scooped: student used AI to derive and post proofs on arXiv first — fortnow · 2026-09-19
- Talk like a real engineer — in systems and tradeoffs — and AI won't replace you — ashishllm · 2026-09-19