Anthropic has agreed to grant researchers from METR, a nonprofit focused on evaluating frontier AI risks, the kind of deep access typically reserved for employees. That means the outside group can examine the company's safety incidents, internal practices, and risk-management decisions firsthand rather than relying on whatever Anthropic chooses to disclose publicly. Crucially, Anthropic has agreed not to exercise any veto power over what METR ultimately publishes.
This marks a significant departure from how AI labs have typically handled outside oversight. Until now, most safety audits and assessments of companies like Anthropic, OpenAI, and Google DeepMind have relied on limited, company-controlled disclosures, voluntary commitments, or narrow pre-release model testing. Giving an external group ongoing, employee-level access — and surrendering the ability to suppress critical findings — is a far more aggressive form of transparency than the industry has generally embraced.
The move comes as pressure mounts on AI developers to prove that safety claims aren't just marketing. Regulators, researchers, and the public have grown increasingly skeptical of self-reported safety metrics from companies racing to build more powerful models. METR has already built credibility in this space through its work estimating how capable AI systems are at autonomously completing complex, multi-step tasks — a key signal used to gauge how much autonomy and potential risk a model might carry.
For Anthropic, which has positioned itself as the safety-conscious alternative among major AI labs, the arrangement doubles as both a credibility test and a bet. If METR's assessments turn up serious problems, Anthropic has committed to taking the hit publicly rather than quietly managing the narrative. If the findings are largely reassuring, the company gains a powerful, independently verified endorsement that rivals can't easily match.
The arrangement could also set a template other AI companies face pressure to match. As governments weigh AI safety legislation and international bodies debate oversight frameworks, a precedent of independent, non-vetoable auditing may become a reference point for what meaningful accountability looks like — or a bar that competitors are pushed to clear, whether they're ready to or not.