Skip to content
Neural Daily

PolicyNews

Anthropic gives METR employee-like evaluator access

Anthropic said on October 15 that six METR employees now have employee-like access to company hardware, office space and internal tools, roughly on par with Anthropic’s own risk teams. METR can publish risk, incident and practice assessments; Anthropic has no veto, only narrow redactions.

By: Mikola Máté Patrik · · 2 min read

Anthropic: An update on independent evaluators
Image: Anthropic · Screenshot · Screenshot of official page · Source

What METR gets

Six staff from the nonprofit METR receive Anthropic hardware, office space and internal systems. Access is described as roughly equivalent to Anthropic’s own risk-assessment teams. Outside narrow categories such as customer data, attorney-client privilege and model architectures, staff are encouraged to share information with them as with a colleague on alignment risk (Anthropic).

METR will publish its analysis of risk levels, incidents or practices. Anthropic cannot veto publication but can require redactions in those predefined categories. METR may characterize redactions publicly and say whether it believes they were justified, without revealing the redacted content, and can challenge them in public.

Three tiers of access

Anthropic outlines three tiers. Tier one: narrowly scoped risk investigations, access decided by Anthropic, publication with approval. Tier two: more information, publication without approval, redactions still allowed. Tier three: the METR-style employee-like arrangement, with publication rights similar to tier two. The goal is several evaluators at tier three over time.

Other partners and context

SecureBio is reviewing the chemical and biological weapons chapter of the August Risk Report after a published pilot. Accenture and Faculty are building toward model evaluation, red-teaming, alignment assessments and safeguard testing. Early talks include Hugging Face. The move sits under the White House Accord on Super Intelligence, where companies pledged to work with independent auditors or evaluators. For a European parallel, see our piece on the EU AI Office information requests.

Key takeaways

  • Six METR staff get employee-like access: hardware, office, internal tools.
  • No publication veto; only narrow redactions (customer data, privilege, architectures).
  • METR can publicly challenge redactions without revealing the content.
  • Three tiers: narrow probes → more info and freer publishing → employee-like.
  • In practice this is voluntary oversight, not a regulatory licence. Analysts argue frontier labs will find it harder to justify having no comparable external view.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles