Anthropic surrenders veto power over external safety audits

The detail that matters most is editorial independence: Anthropic cannot suppress or soften what METR publishes, which is structurally different from the sponsored audits and cherry-picked red-team disclosures that have defined external safety reviews until now. That constraint is what gives this arrangement actual teeth, and it's what prior voluntary transparency gestures have consistently lacked.

This move lands in a week when Anthropic is under simultaneous pressure from multiple directions. The FTC probe reported by Ars Technica on October 15th is investigating whether Anthropic security incidents caused consumer harm, and the false homicide tip to Philadelphia police from October 9th gave regulators a concrete incident to anchor that inquiry. Granting METR unrestricted publishing rights looks less like altruism in that context and more like a calculated effort to demonstrate proactive accountability before regulators impose a less favorable oversight structure. The earlier story on Anthropic isolating its internal evaluations from the internet also matters here: that architectural change was a containment response, while this METR arrangement is a disclosure response. They are solving different parts of the same governance gap.

Watch whether METR publishes a finding that reflects negatively on Anthropic within the next six months. If the first reports are uniformly favorable, the editorial independence claim deserves serious scrutiny. If a critical finding does ship without Anthropic interference, that establishes this as a genuine oversight model other labs will face pressure to replicate.