Anthropic said on 15 October 2026 that it has started giving employee-like access to a team from METR, a nonprofit that evaluates frontier AI models, according to a post on its website. Anthropic says it has onboarded six METR employees with company hardware, office space and access to internal tools and systems roughly on par with what its own risk assessment teams have.
METR will publish its analysis of Anthropic’s risk levels, incidents or practices, the post says. Anthropic says it can’t veto publication, though it can require METR to redact information in narrow, predefined categories.
Business Pill · AI AUDIT
A one-minute explainer of the AI audit problem: why checking AI work can cost more than producing it. It teaches the general idea only and says nothing about any organisation in this story.
The key insight: As we read it, the arrangement moves part of Anthropic’s safety disclosure from the company to an outside team: Anthropic says METR will publish its own analysis of Anthropic’s risk levels, incidents or practices, and that Anthropic can’t veto publication.
According to the post, Anthropic employees are encouraged to share information with the METR team “just as they would with a fellow employee working on alignment risk.”
The post lists a small number of categories that are outside that, giving as examples data about customers, attorney-client privileged information and information about Anthropic’s model architectures.
Anthropic says its engagement with METR is limited to a small team operating under enforceable agreements, and that other METR employees are siloed from that team’s work, including researchers who work with Anthropic on other projects.
The post says embedded evaluation is “not a way for other organizations to obtain our information, or that of our customers and partners.”
Anthropic says it has onboarded six METR employees with company hardware, office space and access to internal tools roughly on par with its own risk assessment teams; METR will publish its analysis and Anthropic can’t veto publication (Anthropic, 15 October 2026).
What Was Posted on X
Weโve committed to having evaluators assess our safety practices. We're in discussions with multiple organizations, and have now given employee-like access to a team from @METR_Evals.
Anthropic’s own post on 15 October 2026 says it has ‘given employee-like access to a team from @METR_Evals’ and links its post describing how it works with third-party evaluators, the primary source for this piece.
๐จAnthropic has embedded six METR employees inside the company, giving them Anthropic hardware, office space and internal access roughly on par with its own risk-assessment teams.
A summary by @choblin29 of Anthropic’s post: six embedded METR staff with Anthropic hardware, office space and internal access, publication without Anthropic approval subject to narrow redactions METR can dispute, and three evaluator tiers. The post also cites a Reuters report from two days earlier on labs seeking control over evaluators’ publications, which we have not read.
Who Decides What Gets Published
Anthropic says that, while it can require redactions in predefined categories, METR is free to characterize the redactions publicly and say whether it believes they were justified, as long as it does not reveal the redacted information.
METR set out its own terms a day earlier. In a 14 October 2026 post, it said it will publicly describe the scope and duration of its work with companies, any access limitations that affect its conclusions, and whether a company redacted information that was important to them.
METR says one arrangement it has found works well is retaining the right to unilaterally and prominently attach a “redaction summary” to its reports.
Three Tiers of Access
Anthropic says it is developing three tiers of access for evaluators. In the first tier, evaluators conduct narrowly scoped risk investigations, Anthropic decides what access to provide, and evaluators can publish their findings with its approval.
In the second tier, the post says, evaluators can publish without Anthropic’s approval and have access to more information, though Anthropic can still require the removal of specific categories of information.
The third tier is the METR arrangement, with employee-like access and publication rights similar to the second tier’s, according to the post.
Anthropic says SecureBio is currently in the first tier and is reviewing the chemical and biological weapons chapter of its August Risk Report. It says it is working with Accenture and Faculty toward an ambitious partnership, and that it is in early discussions with several other organizations, including Hugging Face.
Why Anthropic Says It Is Doing This
The post says Anthropic committed last month to stronger independent, third-party assessment of its models and safety practices, and that it signed the White House Accord on Super Intelligence, in which companies pledged to partner with independent external auditors or evaluators.
Anthropic says it can’t publicly share all the information relevant to its safety work, because some of it could be useful to malicious actors or is commercially sensitive, like compute or training information. Instead, it says, it is giving third-party evaluators deep access.
The post compares the arrangement to bank supervision, where regulators assign resident examiners to work inside the largest banks, while noting that the analogy isn’t exact.
“We’re embedding evaluators voluntarily, and we believe every frontier AI company should do the same,” the post says.
What METR Says About Its Independence
In its 14 October post, METR says it has chosen not to accept payment or donations from frontier AI companies, and that although it accepts a large amount of free model access, it ensures it can afford to pay for any free tokens it receives.
METR also writes that “it has proved challenging to ensure that we have adequate independence to collect and publish evidence under the current voluntary regime.” It says it remains dependent on the companies it works with to voluntarily engage, provide access and permit it to publish.
As we read it, the access is the product: Anthropic says the six METR employees have internal tools and systems roughly on par with its own risk assessment teams, while customer data, privileged material and model architectures stay outside.
As we read it, the redaction rule is where the trust is tested: Anthropic can require redactions in predefined categories, and METR says it will publicly describe whether redacted information was important to its conclusions.
As we read it, the tiers set a ladder other evaluators can climb: Anthropic says first-tier evaluators publish with its approval, second-tier ones without it, and only METR so far has employee-like access.
Anthropic, in its 15 October 2026 post
“We want to be publicly accountable for the decisions weโre making.”
Three Implications
ANTHROPIC’S EVALUATORS Anthropic says SecureBio is in the first tier, that it is working with Accenture and Faculty toward an ambitious partnership, and that it is in early discussions with Hugging Face.
OTHER FRONTIER LABS Anthropic says it believes every frontier AI company should embed evaluators voluntarily, as it has.
METR METR says it remains dependent on the companies it works with to voluntarily engage, provide access and permit it to publish.
The framework argues that Anthropic “is becoming the first major AI lab actively attempting to define the governance layer of the AI stack.”
As we read it, the METR arrangement is a concrete piece of that governance layer: Anthropic sets the access tiers and redaction categories, and says every frontier AI company should embed evaluators. The framework predates the announcement and does not discuss METR.
Anthropic’s 15 October 2026 post: METR is at the third of three access tiers; SecureBio is in the first; Anthropic is working with Accenture and Faculty toward an ambitious partnership and is in early discussions with Hugging Face.
What Is Not Established
The two posts we read give no duration or end date for the METR engagement, no date for METR’s first published assessment under it, and no payment terms.
The posts also do not name which Anthropic models the METR team is assessing.
Business Engineer Framework
Inside Anthropic’s Permission Layer
A Business Engineer framework on Anthropic’s attempt to define the governance layer of the AI stack.
Anthropic says it has embedded six METR employees with employee-like access, roughly on par with its own risk assessment teams, and that METR will publish its analysis of Anthropic’s risk levels, incidents or practices without an Anthropic veto.
Anthropic describes METR as the only evaluator so far at the third of three access tiers, with SecureBio in the first, and says every frontier AI company should embed evaluators. METR says independence remains hard to ensure under the current voluntary regime.
A note on sourcing. We read Anthropic’s 15 October 2026 post “An update on independent evaluators” and METR’s 14 October 2026 post “Our approach to working with AI companies” in full; every term of the arrangement comes from them. The IPO, FTC and third-quarter context comes from our own earlier reporting. We did not use other outlets’ reports.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur, HackerNoon and Search Engine People. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
