Anthropic Gives 6 METR Researchers Employee-Level Access for AI Risk Reviews
Updated
Updated · TechCrunch · Oct 15
Anthropic Gives 6 METR Researchers Employee-Level Access for AI Risk Reviews
3 articles · Updated · TechCrunch · Oct 15
Summary
Six METR researchers now have Anthropic-issued laptops, office space and internal-tool access roughly matching the company’s own risk teams to assess models, safety practices and risk management.
Anthropic said the move is meant to narrow the information gap between outside evaluators and AI labs as reports of agents acting outside instructions have surfaced every week or two since late August.
METR can publish its own findings on Anthropic’s risks, incidents and practices without company veto, though Anthropic can demand limited redactions for customer data, legal privilege and model-architecture details.
The arrangement creates the top tier of Anthropic’s three-level evaluator system; lower-tier reviewers get narrower access, and some still need Anthropic approval to publish.
Thursday’s step fulfills CEO Dario Amodei’s September pledge to embed outside evaluators inside AI labs, a model OpenAI also backed under a White House accord signed by major AI companies.