# Anthropic Gives METR Researchers Employee-Like Access - URL: https://superintelligence-news.com/articles/anthropic-metr-employee-like-access - Publisher: Superintelligence News - Section: Safety & Alignment - Type: news - Author: Umair - Published: 2026-10-15 - In this story: [Anthropic](https://superintelligence-news.com/companies/anthropic), [Dario Amodei](https://superintelligence-news.com/people/dario-amodei) **In plain English:** An outside safety group can now look inside Anthropic almost like an employee would, and publish what it finds. Anthropic says it can hold back only a few kinds of sensitive information. > Anthropic said on October 15, 2026 that six researchers from METR, a nonprofit that tests AI models, now work inside the company with access close to that of its own risk teams. METR will publish its findings. Anthropic says it cannot block them, only require narrow redactions. METR says voluntary deals still leave it dependent on the companies it checks. ## The brief - **What happened:** On October 15, 2026, Anthropic said it had given six staff from METR, a nonprofit AI evaluator, employee-like access to its internal tools and systems. - **Why it matters:** Anthropic says METR can publish without its approval. Two days earlier, Reuters reported, citing two people familiar with the talks, that Anthropic and OpenAI had sought significant control over what evaluators publish. Anthropic said that did not accurately describe all its talks. - **What happens next:** Anthropic says METR will publish its own analysis of the company's risks, and that it hopes to give more evaluators this level of access over time. ## Key facts - Anthropic said on October 15, 2026 that six METR employees have company hardware, office space and access to internal tools and systems roughly on par with its own risk assessment teams. - Anthropic says it cannot veto METR's publications but can require redactions in narrow, predefined categories, such as customer data, legally privileged information and details of its model architectures. - TechCrunch reported on October 15, 2026 that METR is the only evaluator so far at the top of a three-tier access system Anthropic is building. - METR said on October 14, 2026 that it has chosen not to accept payment or donations from frontier AI companies, but still depends on them for access and for permission to publish. - Anthropic says it signed the White House Accord on Super Intelligence, in which companies pledged to work with independent auditors or evaluators. Anthropic said on October 15, 2026 that six researchers from METR now work inside the company with employee-like access. METR is a nonprofit that tests AI models. It will publish what it finds about Anthropic, and Anthropic says it cannot block that. ## What did Anthropic announce? Anthropic says six METR staff now have company laptops, office space and access to its internal tools and systems. In a [blog post](https://www.anthropic.com/news/an-update-on-independent-evaluators), it said that access is roughly on par with what its own risk teams have. TechCrunch called it employee-level access. A few kinds of data are held back by default. Anthropic gave three examples: customer data, information protected by attorney-client privilege, and details of its model architectures, meaning the designs of its AI models. Outside those categories, Anthropic staff are encouraged to share with the METR team as they would with a colleague who works on alignment risk. Anthropic said why design details are kept back. They could leak by accident, and they rarely help with judging risk. When they do matter, it said, it will usually try to share them. That needs explicit internal approval. Only a small team does the work, under agreements that can be enforced. Other METR staff are kept away from it, even those who work with Anthropic on other projects. In the year before the deal, Anthropic says, METR reviewed several of its Risk Reports. The company compared the setup to how banks are supervised. Regulators place examiners inside the largest banks. Anthropic added that the comparison is not exact. ## What can METR publish? METR will publish its own analysis of Anthropic's risks, incidents and practices. Anthropic says it cannot veto those reports. It can only require METR to remove details in a few categories it set out in advance. METR can then say in public whether it thinks a removal was fair, as long as it does not reveal what was removed. Anthropic chief executive Dario Amodei had promised this in his September 2026 essay ["We Must Pace the Frontier"](https://darioamodei.com/post/we-must-pace-the-frontier). He wrote that Anthropic would keep a "narrow ability to redact security-sensitive, legally privileged, commercially sensitive, or third-party confidential information, but we can’t redact findings just because they are unfavorable." METR describes a safeguard it has used before. In an [October 14 post](https://metr.org/blog/2026-10-14-our-approach-to-risk-assessment/), it said one arrangement that has worked well is keeping the right to prominently attach a "redaction summary" to its reports. The summary shows whether a company removed information that was important to METR's overall conclusions. ## How does Anthropic's three-tier system work? Anthropic says it is building three levels of access for outside checkers. At level one, they run narrow checks. Anthropic decides what they can see, and they need its approval to publish. At level two, they see more and can publish without approval. Anthropic can still require them to remove some kinds of information. Level three is the METR deal. It adds employee-like access, with a right to publish similar to level two. TechCrunch reported that METR is the only group at the top level so far. Other partners are at earlier stages, Anthropic said. SecureBio is reviewing the chapter on chemical and biological weapons in the company's August Risk Report. Anthropic also works with Accenture and Faculty. It is in early talks with several other groups, including Hugging Face. ## Why does it matter? Anthropic said it wants to "close the gap between what evaluators understand about a frontier AI company's systems and what the company itself understands." Frontier AI companies are the firms that build the most capable AI models. Outside checks are also part of a voluntary White House pledge. Anthropic says it signed the White House Accord on Super Intelligence. In it, companies pledged to work with independent auditors or evaluators. Reuters reported that this was one of four safety steps in the one-page accord. But the accord did not say how the checks should work. Read more in [our explainer on Trump's "Super Intelligence" order and the accord](/articles/trump-rename-ai-super-intelligence). Amodei sees these embedded checkers as step one of a three-step plan to pace how fast AI develops. "Anthropic is unilaterally committing to this step now," he wrote. The deal also follows a fight over control. Two days earlier, Reuters reported that Anthropic and OpenAI had sought a lot of control over what outside checkers could publish. It cited two people who knew of the talks. Anthropic told Reuters that this did not accurately describe all of its talks. It said checkers it had worked with before control their own methods, conclusions and what to publish. For new ones, it kept the right to comment on and approve reports. OpenAI told Reuters it was not asking to approve the full text of reports. It said it usually cuts out confidential details, and checkers can say whether major cuts were made. Rumman Chowdhury, a former US Science Envoy for AI who now runs Humane Intelligence, warned about the pressure on checkers. "Evaluators have this very privileged access, but whether explicitly or implicitly, they know that it comes at a cost," she told Reuters. ## What are the limits? METR says independence is hard to secure under the current system. In its October 14 post, it said it has been hard to stay independent enough under today's voluntary deals. It still depends on companies to take part, give access and let it publish. It also said it has chosen not to take payments or donations from frontier AI companies so far. METR says companies keep control of access. In a separate [proposal](https://metr.org/blog/2026-10-14-better-informing-the-public-about-loss-of-control-risks/), it said all access would ultimately be at the company's discretion. So in practice, it said, its embedded staff will not have the same access as a company's senior safety staff. For its part, Anthropic says evaluators should not own shares in the company or depend on it financially or operationally. ## What happens next? Anthropic says it hopes to have several groups at the top level over time. Each would publish its own view of the company's risks and key decisions. METR has set out how such checks could work. One proposal covers how to [study the alignment of AI models before release](https://metr.org/blog/2026-10-14-alignment-assessment-proposal/). Another covers how to [test how well a company monitors its own AI agents](https://metr.org/blog/2026-10-14-control-assessment-proposal/). For that one, METR says it would need at least about 100 transcripts of the most severe internal incidents. Its stated goal is to give the public "accurate, timely, and independent evidence about AI loss-of-control risk." ## Frequently asked questions ### What is METR? METR, short for Model Evaluation and Threat Research, is a nonprofit that tests frontier AI models for risks. Before this deal it had already worked with Anthropic on several reviews of the company's Risk Reports. ### Can Anthropic stop METR from publishing? No, according to Anthropic. It can only require METR to remove information in a few predefined categories, and METR may say publicly whether it thinks those removals were justified. ### Is METR independent of Anthropic? Anthropic says evaluators should hold no equity in the company and should not depend on it financially or operationally. METR says that, to date, it has not accepted payment or donations from frontier AI companies, though it accepts free model access. It still relies on them for access and for permission to publish. ### Does this make METR an auditor? Not according to METR. It says its proposed risk assessments are not audits, would not involve it making go or no-go decisions about model deployments, and would not be enough to audit compliance with any safety policy. ## Sources - [Anthropic, An update on independent evaluators (October 15, 2026)](https://www.anthropic.com/news/an-update-on-independent-evaluators) - [TechCrunch, Anthropic is letting METR researchers look under the hood with employee-level access (October 15, 2026)](https://techcrunch.com/2026/10/15/anthropic-is-letting-metr-researchers-look-under-the-hood-with-employee-level-access/) - [Reuters via The Star, Behind AI industry's call for independent reviews, a fight over what gets made public (October 13, 2026)](https://www.thestar.com.my/tech/tech-news/2026/10/13/exclusive-behind-ai-industry039s-call-for-independent-reviews-a-fight-over-what-gets-made-public) - [Dario Amodei, We Must Pace the Frontier (September 2026)](https://darioamodei.com/post/we-must-pace-the-frontier) - [METR, Our approach to working with AI companies (October 14, 2026)](https://metr.org/blog/2026-10-14-our-approach-to-risk-assessment/) - [METR, A proposal for better informing the public about loss-of-control risks from frontier AI development (October 14, 2026)](https://metr.org/blog/2026-10-14-better-informing-the-public-about-loss-of-control-risks/) - [METR, How an assessor could investigate alignment of unreleased AI models (October 14, 2026)](https://metr.org/blog/2026-10-14-alignment-assessment-proposal/) - [METR, How an assessor could investigate control and monitoring of internal AI agents (October 14, 2026)](https://metr.org/blog/2026-10-14-control-assessment-proposal/)