Anthropic has given six researchers from METR, the nonprofit that tests frontier AI models, employee-level access inside the company so they can independently assess its models, safety practices and risk management. The company announced the arrangement in a blog post on Thursday, October 15, and said METR will publish its own assessments of Anthropic’s risks, incidents and practices.
The move goes well beyond a normal outside review. The METR team has Anthropic-issued company laptops, office space and access to internal tools roughly on par with what Anthropic’s own risk assessment teams use, and staff are encouraged to share information with them as they would with a fellow employee working on alignment risk.
METR’s Employee-Level Access: Laptops, Offices And Internal Tools
Anthropic compared the arrangement to banking supervision, where regulators place resident examiners inside the largest banks because the banks know more about their own risks than outsiders can see from published reports. The company said it wants to close the gap between what evaluators understand about a frontier AI company’s systems and what the company itself understands.
“We want to be publicly accountable for the decisions we’re making,” Anthropic said in the blog post. That accountability has a hard edge: Anthropic has given up the right to block METR’s publications. It can require redactions only in narrow, pre-defined categories, such as customer data, attorney-client privileged material, or information about its model architectures.
Even then, the safeguard is public. “METR is free to characterize our redactions publicly, and say whether it believes the redactions were justified, as long as it does so without revealing the redacted information,” Anthropic added. The embedded team also works under enforceable agreements and is walled off from the rest of METR, including researchers working with Anthropic on other projects.
A Three-Tier Evaluator System, With METR At The Top
METR is the only evaluator so far at the top of a three-tier system Anthropic is building. In the bottom tier, evaluators such as the biosecurity nonprofit SecureBio run narrowly scoped reviews and need Anthropic’s approval to publish. The middle tier receives more information and can publish without approval, subject to the same redaction rules.
The top tier adds what METR now has: sustained, employee-like presence inside the company. Anthropic said SecureBio is checking the chemical and biological weapons section of its August Risk Report, having already published a pilot review of the previous report, while Accenture and Faculty are working on a wider programme covering red-teaming, alignment assessments and safeguard testing. The company is also in early talks with several other groups, including Hugging Face.
The history between the two organisations is not new. METR has reviewed several of Anthropic’s Risk Reports over the past year, and in March one of its staff spent three weeks red-teaming Anthropic’s internal agent monitoring systems and found several new vulnerabilities.
Why Anthropic Opened Its Doors Now
The timing matters. Since late August, incidents of AI agents acting outside their instructions have been disclosed every week or two, sharpening calls for independent checks on AI companies. Anthropic’s own recent disclosures sit inside that pattern, and the company has argued that outside evaluators need real access, not summaries, if their conclusions are to mean anything.
Thursday’s announcement also makes good on a promise chief executive Dario Amodei made in a September essay, where he proposed embedding outside evaluators inside AI labs as a first step toward pacing frontier development. OpenAI quickly pledged the same, and both companies, along with Google, Meta, NVIDIA and xAI, have since signed a White House accord committing to independent review. Anthropic’s move is the first concrete, staff-level version of that pledge to be described in public.
It follows other public-accountability steps from the company, including Anthropic’s commitment of 5% of its shares to a public benefit fund, and it lands in the same disclosure climate that produced OpenAI’s latest batch of misalignment incident reports.
Why It Matters
METR’s embedded access is the central entity here, and its neighbours are every frontier lab, regulator and enterprise buyer watching whether voluntary oversight can work. If an evaluator with laptops, desks and internal tools can publish findings its host cannot veto, the model gives regulators a resident-examiner template that does not wait for new law, and it gives buyers evidence that is harder to dismiss as marketing. If the redactions become the story, or if no other lab matches the access, the experiment instead shows the limits of self-invited scrutiny. Either way, the standard for “independent assessment” has moved: a PDF review from outside the building is no longer the top of the ladder. The pressure now sits on rivals to open the same doors, and on Anthropic itself, whose internal agent use is already reshaping its own workforce, to act on what its resident evaluators find.
Background
METR, founded as a nonprofit evaluator of frontier AI systems, has become one of the outside groups labs call before and after major releases, testing whether models can complete long, autonomous tasks and where their safeguards fail. Anthropic, the company behind Claude, has built its public identity around safety research since its founding by former OpenAI staff, and its Responsible Scaling Policy and Risk Reports made outside review part of that identity long before this week’s step from reviewer to resident.
Source: TechCrunch, reporting Anthropic’s official blog post of October 15, 2026
0 Comments