METR will publish independent assessments of Anthropic's risks and incidents, and Anthropic has given up the right to block the findings.
Anthropic is granting METR, an external AI safety research group, employee-level access to its internal systems. METR will produce its own public reports on Anthropic's risk practices and incident handling, without Anthropic retaining veto power over what gets published.
This goes further than typical third-party audits or red-teaming arrangements, which usually involve NDAs and company-approved disclosure. It sets a transparency bar that regulators and enterprise buyers can point to when evaluating other frontier labs.
Enterprise customers buying AI for regulated workflows need provable oversight, not vendor self-attestation. If this arrangement holds and produces credible reporting, it becomes a procurement differentiator and a template competitors will face pressure to match.
The daily signal, curated. Get it in your inbox.
Subscribe on LinkedIn →