BusinessClaude

Anthropic gives METR employee access to audit models

Original title: Anthropic is letting METR researchers look under the hood with employee-level access

Plain language

What it is: Anthropic is giving six researchers from the nonprofit METR employee-level access to evaluate its AI models and safety practices. METR will be allowed to publish its findings without Anthropic having the right to block them.
How it works: The researchers receive Anthropic laptops, office space, and access to internal tools. Anthropic employees are encouraged to share information with them just like they would with co-workers.
Where it is available: This arrangement applies to Anthropic's internal systems, models, and risk management practices.
Example: METR will use this access to write and publish independent assessments of Anthropic's risks, incidents, and safety practices.

What it is: Anthropic has granted six METR researchers employee-level access to independently assess its frontier models and risk management. METR has the right to publish its assessments, and Anthropic cannot veto publication except for narrow redactions like customer data or model architecture.
How it works: METR researchers are embedded within Anthropic with company laptops, office space, and access to internal risk assessment tools. They can communicate with Anthropic employees on alignment risks.
Where it is available: This is part of a new three-tier evaluation system, with METR occupying the top tier with full employee-level access.
Example: METR will conduct independent evaluations of Anthropic's models and publish reports on safety incidents and practices.

The headline and explanation on this page are based on TechCrunch's post. The full details are in the original.

Anthropic is letting METR researchers look under the hood with employee-level access