Anthropic gives METR evaluators employee-level access to its AI models

By Splitfeed EditorialLast updated Oct 15, 2026, 18:40 UTC

On October 15, 2026, Anthropic announced that it had granted employee-level access to six evaluators from METR, a nonprofit organization focused on AI safety. The arrangement allows those evaluators to conduct safety reviews from inside Anthropic's systems, giving them visibility into frontier models that goes beyond what standard external access typically provides. Anthropic framed this as an update to its independent evaluator program, with the onboarding of the six METR staff representing the first concrete step under the new arrangement.

The significance of this move lies in what it signals about the structure of AI oversight. Independent evaluations of frontier models have historically been constrained by the access companies are willing to grant, which critics have argued limits what outside reviewers can actually find. By offering access comparable to that of its own employees, Anthropic is extending a level of transparency that sets a higher bar than most voluntary arrangements in the industry. The genuine point of dispute is whether this kind of company-controlled access, however deep, constitutes meaningful independent oversight, or whether it remains fundamentally limited by the fact that Anthropic itself determines the terms and scope of the arrangement.

None of the three camps had published reactions at the time of writing, so what follows reflects what each would typically argue about a story of this kind. The Pro-AI camp would likely treat this as evidence that leading AI developers are self-correcting and capable of building accountability into their own processes without regulatory mandates. The Anti-AI camp would be expected to argue that access granted voluntarily by the company being evaluated cannot substitute for independent oversight with real authority, and that the arrangement leaves Anthropic in control of what METR can see and report. The Middle Ground camp would probably welcome the step as a genuine improvement while pressing for clearer public reporting requirements and asking whether METR can publish its findings without restriction.

The details most worth watching are what METR is permitted to disclose from its evaluations, whether Anthropic commits to acting on findings, and whether other frontier labs adopt similar arrangements, which would indicate whether this becomes an industry norm or remains a one-off gesture.