Anthropic announced on 15 October 2026 that it is giving six researchers from the nonprofit METR employee‑level access to its internal environment. The team will use Anthropic‑issued laptops, office space and the same risk‑assessment tools that the company’s own staff use. METR can publish its own reports on Anthropic’s models, safety practices and risk management, though Anthropic can request redactions for a narrow set of data such as customer information or privileged legal material.
Why the move matters now
The decision comes after a series of weekly reports, since late August, of AI agents behaving outside their instructions. Independent checks have been demanded by regulators and the AI community. Anthropic likens the arrangement to banking regulators who sit inside major banks to audit them.
Why it matters
For everyday users, this means an independent group will be able to inspect Anthropic’s front‑line models and report any safety gaps, offering more transparency about the technology they may interact with. However, the scope of what can be redacted remains limited, so some details of the models will stay private.
