The Pulse
Anthropic Gives Six METR Evaluators Employee-Like Access
Anthropic says six METR employees now have company hardware, office space and access to internal tools for independent risk assessments. METR can publish its findings, though Anthropic retains limited redaction rights and excludes some sens
AI.info Team · Written with AI ·
METR warned on October 14 that companies could limit what outside evaluators see; the next day, Anthropic said it had given six METR employees employee-like access to its internal systems. The arrangement gives the nonprofit a closer view of the company’s safety work and the right to publish its assessments, but it leaves Anthropic authority to withhold certain information and require narrow redactions.
Six evaluators, with access to internal tools
Anthropic says METR’s six-person team has company hardware, office space and access to internal tools and systems roughly comparable to what Anthropic’s own risk assessment teams use. The company says employees are encouraged to share information with the evaluators as they would with colleagues working on alignment risk.
Anthropic excludes a small set of categories, including customer data, attorney-client-privileged information and information about its model architectures. The METR team operates under enforceable agreements, and other METR employees—including researchers working with Anthropic on separate projects—are kept apart from the embedded team’s work.
The company says METR can publish assessments of its risks, incidents and practices without Anthropic’s approval. Anthropic can require redactions for the defined sensitive categories, while METR can publicly describe those redactions and say whether it considers them justified, without disclosing the withheld material.
METR’s warning about company-controlled access
A day before Anthropic’s announcement, METR described the limits that can come with voluntary assessments. “Since access requests are approved or denied at the company’s discretion, embedded assessors from METR will not have the same degree of access as a company’s senior safety staff,” the nonprofit wrote in its proposal for public risk assessments.
METR also said it could not offer confident assurance that AI development is safe: the science is not mature enough for consensus conclusions, it wrote, and there are no standards against which to audit safe development. Its proposed role is to gather evidence for the public, not decide how companies should develop or deploy AI.
Anthropic’s announcement describes a more formal arrangement than a limited external review, but it does not remove all access constraints. The company’s own description rules out some information and limits the embedded team to six people working under its security standards. Neither the announcement nor METR’s proposal establishes that the evaluator can inspect every system or document relevant to a safety question.
METR joins a three-tier evaluation plan
Anthropic places the METR arrangement in a third tier of evaluator access. In the first tier, Anthropic decides what information an evaluator receives and must approve publication. Second-tier evaluators get broader access and can publish without approval, subject to specific redactions; Anthropic describes METR’s arrangement as combining employee-like access with publication rights similar to that second tier.
The company says it has also worked with METR on several reviews of its Risk Reports over the past year. It names SecureBio and Accenture as other evaluators, while saying those partnerships remain at earlier stages with more limited information and publication rights. Anthropic also says it is in early discussions with organizations including Hugging Face.
Publication rights are the test ahead
Anthropic says it wants several evaluators eventually operating at the third tier, each publishing its own analysis of the company’s risk posture and key decisions. The announcement does not give a schedule for METR’s first report under the new arrangement or specify how long the embedded access will last.
The test will be whether METR can publish findings that show both what it examined and what it could not examine. Anthropic says the evaluator can describe its redactions and challenge their justification publicly; METR has separately said that public reports should disclose access limits and other constraints. The company announced the arrangement on October 15, 2026, as part of its broader update on independent evaluators.