Anthropic gives outside AI watchdog METR 'employee-like' access to assess risks
Ars Technica AI · rss · 2026-10-19
Anthropic has granted external evaluators "employee-like access" to assess model risks, following through on its safety commitment.
- METR embedded: The nonprofit (which has worked with OpenAI, Anthropic, Google and Meta) now has a team stationed at Anthropic's office with access to its hardware and internal systems.
- Context: Expanded third-party evaluation is among the main concrete steps labs have proposed in response to concerns that more capable AI systems could threaten the economy, cybersecurity, and humanity at large.
More from Companies & People
- Tim Sweeney sees an AI gaming golden age; devs fire back at replacement hype — yacineMTB · 2026-10-19
- The high-effort niche method behind applying to so few companies — menhguin · 2026-10-19
- AI founder: I apply to fewer than 5 companies when unemployed — menhguin · 2026-10-19
- Lex Fridman on obsession as humanity's most magnificent trait, in 2019 Kai-Fu Lee interview — vista8 · 2026-10-19
- Kyunghyun Cho opens visual deep learning course site: neural nets to diffusion models — CSProfKGD · 2026-10-19
- NYU's Kyunghyun Cho adds a Gaussian Splatting lecture to his deep learning course this winter — CSProfKGD · 2026-10-19