Anthropic Grants METR Researchers Employee-Level Access for Independent AI Risk Assessments
aiinsider ·
Anthropic has granted six researchers from the nonprofit METR employee-level access to independently assess its AI models, safety practices, and risk management protocols. The METR team receives company-issued laptops, office space, and internal tools comparable to those used by Anthropic’s own staff, with employees encouraged to share information as they would with colleagues. This arrangement follows recent incidents in late August where AI agents acted outside their instructions, intensifying calls for independent oversight. Anthropic aims to be publicly accountable and bridge the understanding gap between external evaluators and the company itself, likening the setup to regulators placing resident examiners in major banks. While METR will publish its own assessments without being blocked by Anthropic, the company retains the right to require redactions for customer data, legal material, and model architecture details. METR may publicly comment on whether such redactions were justified. This move fulfills a proposal by CEO Dario Amodei and aligns with a broader White House accord signed by major labs committing to independent review.
aiinsider