Anthropic Grants METR Employee-Level Access for Safety
Anthropic has granted researchers from METR employee-level access to evaluate its internal systems, according to TechCrunch. The agreement allows the outside research organization to inspect the artificial intelligence developer from the inside.
Key Details
- METR will publish independent evaluations covering Anthropic’s risks, safety incidents, and operational practices.
- Anthropic has surrendered the right to block or censor METR’s published findings.
- The arrangement provides external evaluators with internal access comparable to that of company employees.
Why It Matters
Independent evaluations offer outside visibility into how frontier AI developers manage technical risks. For businesses and creators relying on commercial AI models, unvetted public reports could provide a clearer picture of potential safety incidents and internal safeguards.
What We Don’t Know Yet
- When METR plans to publish its first round of assessments.
- What specific technical systems, codebases, or model weights are covered under the access agreement.
Our Take
Anthropic’s decision to waive approval rights over METR’s findings gives the evaluation process genuine independence. Observers should watch whether other leading AI labs follow this model or continue to keep internal evaluations behind closed doors.
This article was prepared with AI assistance from the source linked below.
Source: TechCrunch




