Anthropic Lets Outside Watchdogs Look Inside Its AI
This could make AI safer for everyone—and set a precedent for the whole industry.
Anthropic announced Thursday it will give six evaluators from METR, a third-party nonprofit that tests frontier models, “employee-like access,” including Anthropic hardware, office space, and internal tools and systems “roughly on par” with what its own risk assessment teams can access. Outside a small number of clearly defined categories—such as customer data, attorney-client privileged information, or model architecture information—Anthropic employees are encouraged to share information with them “just as they would with a fellow employee working on alignment risk.” METR will publish its findings, none of which Anthropic is allowed to withhold, though Anthropic will ask METR to redact sensitive information where necessary.
- Anthropic is letting outside auditors from METR work inside the company with almost the same access as employees, a first for the AI industry.
- The auditors will publish their findings, and Anthropic can't stop them, though it can ask to redact sensitive information.
- This is voluntary—no law requires it—but it could pressure other AI companies to be more transparent about safety.
Why It Matters
More independent scrutiny of AI could lead to safer, more trustworthy AI tools for everyday use.