Anthropic gives METR employee-like access to models and safety info; METR can publish without veto
Anthropic has onboarded six METR employees with internal hardware, tools, and access on par with its own risk assessment teams, per its update on independent evaluators. METR can publish risk assessments without Anthropic veto, though narrow redaction categories apply. SecureBio and Accenture are also evaluators, with early talks with Hugging Face. Andrew Curran suggests such posts precede major model releases.
All posts · 2, most important first
- We’ve committed to having evaluators assess our safety practices. We're in discussions with multiple organizations, and have now given employee-like access to a team from @METR_Evals. In this post, we describe how we work with third-party evaluators: anthropic.com/news/an-update…138 likes · 14 reposts
- METR now has employee level access to Anthropic's frontier models. Also, this is exactly the type of blog post you make just before releasing a shockingly powerful and capable new model, I'm just saying.
@anthropicai: We’ve committed to having evaluators assess our safety practices. We're in discussions with multiple organizations, and have now given employee-like access to a team from @METR_Evals. In this post, we describe how we work with third-party evaluators: anthropic.com/news/an-update… open on X297 likes · 15 reposts
@anthropicai org · 
@andrewcurran_ commentary ·