Dario's 'Pace the Frontier' essay: Anthropic commits to permanent third-party evaluator access
ajratner · x · 2026-09-14
Dario Amodei published a new essay, 'We Must Pace the Frontier,' arguing the AI industry should slow down, with a three-part plan. Anthropic is unilaterally committing to the first step: giving third-party evaluators permanent, employee-level access to its systems to verify safety measures, report incidents, and assess alignment during training.
Stanford's ajratner amplified a response endorsing the call for 'a much broader and more ingenious stable of evaluations,' arguing this requires eval pluralism: more benchmarks (covering environment complexity, output complexity, and autonomy horizon), more evaluators including subject-matter experts rather than a small group of AI researchers, and open eval infrastructure.
More from Safety
- Sanders proposes 20-year prison term for developers advancing superintelligence — Kyrannio · 2026-09-14
- Activist dressed as Doctor Doom crashes Seattle council to protest surveillance camera expansion — jonerp · 2026-09-14
- Stolen METR API key burned ~$600,000 in credits over three weeks; critics mock the AI-safety eval org — AlexTensor · 2026-09-14
- Epoch AI researcher maps the levers for pacing AI: RSI speed limits, compute caps, agent budgets — connoraxiotes · 2026-09-14
- Altman, Amodei, Hassabis and Musk agree to 'pace the frontier' as AI safety funding and hiring surge — Astral Codex Ten · 2026-09-14
- The Decompilation Book: a free tutorial series teaching how decompilers work, built around tiny-dec — blaizedsouza · 2026-09-14