Anthropic has opened its doors to external reviewers, granting them “employee‑like access” so they can evaluate the risks posed by the company’s models. The move fulfills a promise to tighten safeguards amid fears that advanced AI could spiral out of control.
The nonprofit METR, which has previously worked with leading AI labs such as OpenAI, Google, Meta and Anthropic, now has a team stationed inside Anthropic’s headquarters. This team can interact with the company’s hardware and internal systems as part of its assessment.
Increasing the involvement of third‑party evaluators is one of the key concrete actions that AI labs have put forward in response to growing concerns about the potential economic, cybersecurity and existential threats posed by more powerful artificial intelligence.