Ex-OpenAI researcher Daniel Kokotajlo slams Anthropic over redacted safety evals
connoraxiotes · x · 2026-11-06
Daniel Kokotajlo (ex-OpenAI, author of AI 2027) publicly criticizes Anthropic's heavy redactions around METR evals: the company cites "others could go faster" to justify withholding safety-critical information while not itself slowing down or backing frontier-pacing efforts. He notes Anthropic has itself said human extinction is plausible, yet invokes "commercial sensitivity." His message to Anthropic staff: your company is recklessly gambling with our lives, and judging by actions its similarity to OpenAI and Google outweighs the differences.
More from Companies & People
- Eigen AI founder on inference economics: cheaper tokens expand the whole market — songhan_mit · 2026-11-06
- Prepack Wasn't Dropped for Being Too Hard—Its Output Was Never Worth It — cnakazawa · 2026-11-06
- Greg Brockman: AI to drive major biology, health breakthroughs in 12-18 months — ChrisGPT · 2026-11-06
- Figma and Atlassian VPs take the UXDX stage: what happens to product teams when everyone can build? — davidhoang · 2026-11-06
- Caltech's Van Valen Lab ditches X, publishes all AI models on Hugging Face — anshulkundaje · 2026-11-06
- NVIDIA partners with Box: Jensen Huang showcases Nemotron-powered enterprise AI agents at BoxWorks 2026 — NVIDIAAI · 2026-11-06