Anthropic · Friday, October 16, 2026
Anthropic gives outside testers employee-like access
The File desk · Oct 16, 2026, 11:00 AM UTC
Status
Confirmed as Anthropic’s post on Oct. 15, 2026. The access, the headcount, and the publication rules are the company’s account. This desk did not speak to METR.
- Confirmed
Anthropic said it has started giving employee-like access to a team from METR, a nonprofit that evaluates frontier models, and that METR will publish its risk assessments.
- Confirmed
The company said it onboarded six METR employees with company hardware, office space, and internal tools roughly on par with its own risk teams.
- Confirmed
It said it is also working with SecureBio and with Accenture and Faculty, and is in early talks with other organizations, including Hugging Face.
Anthropic said Thursday it has brought in six people from METR, a nonprofit that tests frontier models, and given them access close to what its own risk staff have. METR is to publish what it finds. Anthropic said it cannot veto publication, but it can require cuts of customer data, privileged legal material, and model-architecture details. The company also named SecureBio, Accenture, and Faculty as evaluators, and said talks with others, including Hugging Face, are early.
The post is dated Oct. 15, 2026. Anthropic said that last month it committed to stronger outside assessment, and that it signed what it calls the White House Accord on Super Intelligence, in which companies pledged to partner with independent auditors or evaluators. This desk did not open that accord.
The six METR employees get company hardware, office space, and internal tools. Anthropic says staff should share information with them as they would with a colleague, outside a few closed categories: customer data, attorney-client material, and information about model architectures. METR can say publicly whether it thinks a required cut was justified, as long as it does not reveal the cut material.
Anthropic describes three tiers. The first is a narrow investigation, and publication needs the company’s approval. The second allows publication without approval, with more information, and the company can still require specific categories to be removed. The third is the METR arrangement: employee-like access, with publication rights like the second tier.
SecureBio is reviewing the chemical and biological weapons chapter of Anthropic’s August risk report, after a published pilot review of an earlier report. With Accenture and Faculty, Anthropic says it is working toward testing models, alignment checks, and safeguard tests. Those other partnerships, the company says, still have more limited information and more limited rights to publish.
This is an access arrangement. It is not a new model.
What is still unknown or disputed
- METR has not, in any page this desk opened, confirmed the headcount or the access terms. The account is Anthropic’s.
- The White House accord was not opened. The description of the pledge is Anthropic’s sentence.
- A footnote says the SecureBio sentence was updated on Oct. 15. This desk did not compare it with an earlier version.
Primary sources
Every claim in this story is drawn from the documents below. If a fetch failed, that is recorded on the card.
Source 1
An update on independent evaluators
Anthropic · October 15, 2026
Anthropic said it has started giving employee-like access to a team from METR and has onboarded six METR employees with company hardware, office space, and internal-tool access roughly on par with its own risk teams. METR will publish risk-assessment results. Anthropic also named SecureBio, Accenture, and Faculty, and early discussions including Hugging Face.
https://www.anthropic.com/news/an-update-on-independent-evaluators