Anthropic grants METR employee-level access for AI safety review — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic is granting six METR researchers employee-level access — including company laptops, office space, and internal tools on par with its own risk assessment teams — and is encouraging employees to share alignment-risk information with them as if they were colleagues.
- METR can publish its assessments of Anthropic's risks, incidents, and practices without company approval; Anthropic retains only the right to require redactions in narrow categories like customer data, attorney-client privileged material, and model architecture details.
- METR is the sole evaluator at the top of a three-tier system Anthropic is building, with the bottom tier (including biosecurity nonprofit SecureBio) requiring Anthropic's sign-off to publish and a middle tier allowed to publish without approval.
- Anthropic drew a banking analogy, likening the arrangement to regulators placing resident examiners inside the largest banks, and tied the move to a surge of weekly AI-agent misbehavior disclosures since late August.
- Anthropic is delivering on CEO Dario Amodei's September essay calling for outside evaluators embedded inside AI labs, a step OpenAI quickly pledged to mirror; both companies plus Google, Meta, Nvidia, and xAI have since signed a White House accord on independent review.
Why it matters: Anthropic is ceding meaningful visibility to an outside nonprofit at a moment when AI-agent incidents are surfacing roughly every week or two, giving METR — and by extension the public — an unfiltered window into the lab's risk practices that no other evaluator currently has. The banking-style embedded model sets a new floor for what 'independent review' means in the industry, and the narrow redaction categories mean METR retains strong publishing leverage.
Ask SkimNews
