Article 281: Anthropic Gave an Outside Evaluator Employee-Level Access - and No Veto

Anthropic has granted six researchers from METR, the nonprofit that tests frontier models, access roughly on par with its own internal risk teams: company laptops, office space, and internal tools. Anthropic says its employees are encouraged to share information with them as they would with a colleague working on alignment risk, and that it has given up the right to block publication. The company can require redaction only in narrow pre-defined categories - customer data, attorney-client privilege, model architectures - and METR is free to characterise those redactions publicly, including saying whether it thinks they were justified.

Key Takeaway: The mechanism is borrowed wholesale from banking: resident examiners placed inside the largest institutions. What matters is where the accountability sits. A lab that cannot veto a report has handed a third party the ability to publish findings it may disagree with, and has done so while incidents of agents acting outside their instructions have been disclosed roughly every week or two since late August. This is the first concrete instance of the "outside evaluators embedded in labs" idea that Anthropic's CEO proposed in September and OpenAI quickly echoed. The open question is whether employee-level access produces employee-level scrutiny, or simply a very well-briefed house guest.

Sources: Anthropic, TechCrunch, Reuters, Financial Times, 15-16 October 2026