Anthropic is letting METR researchers look under the hood with employee-level access
TechCrunch Aditya Mehta ● Covered by 2 sources
Anthropic gave six METR researchers employee-level access to its AI work. It’s meant to make outside checks less fake and more like being inside the room.
Based on reporting by TechCrunch, Aditya Mehta — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic is opening its doors wider to outside scrutiny. Six researchers from METR, a nonprofit that tests frontier AI models, now get employee-level access so they can assess Anthropic’s models, safety practices, and risk management from the inside, the company said Thursday.
This is not a tour pass. The METR team has Anthropic-issued laptops, office space, and access to internal tools that are roughly on par with what Anthropic’s own risk assessment teams use. Anthropic says its employees are encouraged to share information with them the same way they would with a coworker focused on alignment risk. The company is trying to narrow the gap between what an evaluator can see and what the lab itself knows.
The timing is doing a lot of work here. Incidents of AI agents acting outside their instructions have been surfacing every week or two since late August, and that has sharpened calls for independent checks on AI companies. Anthropic is leaning into that pressure rather than pretending it doesn’t exist. It even compares the setup to banking, where regulators place resident examiners inside the biggest institutions.
METR will publish its own assessments of Anthropic’s risks, incidents, and practices, and Anthropic cannot block publication. The company can only require redactions in narrow, pre-defined cases, including customer data, attorney-client material, and information about its model architectures. METR can also say publicly whether it thinks those redactions were justified, so long as it does not reveal the hidden material.
Anthropic says METR is the only evaluator at the top of its three-tier system for outside review. Lower-tier evaluators, such as the biosecurity nonprofit SecureBio, do narrower reviews and need Anthropic’s approval to publish. The middle tier gets more information and can publish without approval, while the top tier gets the deepest access. The move also follows a promise Anthropic CEO Dario Amodei made in a September essay about embedding outside evaluators inside AI labs as a first step toward slowing the technology’s development. OpenAI quickly backed the idea, and Anthropic, OpenAI, Google, Meta, Nvidia, and xAI later signed a White House accord on independent review.
My take — AI-written commentary, not fact-checked reporting
This is the rare AI governance move that looks harder to fake than to announce, which already puts it ahead of most lab self-policing theater. The catch is obvious: if resident examiners become a PR badge instead of a real constraint, the whole thing turns into a very expensive office chair. Still, the industry keeps insisting it can be trusted with itself, so outside people in the building is the bare minimum, not the victory lap.
Read more about this at: TechCrunch
