Anthropic gives AI evaluator METR employee-level access to its models and safety systems
Anthropic has given six METR employees access to its internal systems and risk assessment tools, allowing them to independently evaluate safety practices and publish findings without the company's prior approval.
A team of non-profit AI evaluators from METR has been given employee-level access to Anthropic's systems. The group can independently access its advanced models, safety practices, and risk management processes. This announcement follows concerns among researchers and evaluators that leading AI labs may wield too much influence over independent reviews through publication restrictions and control over access to sensitive information.
Six METR employees have been provided with company hardware, office space and access to internal tools and systems broadly comparable to those available to Anthropic's own risk assessment teams, the company said.
Anthropic added that METR will be able to publish independent assessments of risks, incidents and safety practices at the AI lab without prior approval, subject only to limited redactions covering areas such as customer data, privileged information and architectural details.
Access for evaluators
In its own blog post, Anthropic said six METR employees have been given company hardware, office space and access to internal tools and systems on a par with those available to its own risk assessment teams. Outside a small number of clearly defined categories, Anthropic employees are encouraged to share information with them as they would with a fellow employee working on alignment and risk.
METR will be able to publish its analysis of risks, incidents and practices without Anthropic being able to veto publication. Anthropic further noted that its engagement with METR is limited to a small team operating under enforceable agreements. Other METR employees are siloed from that team's work, including researchers who work with Anthropic on other projects.
Tech
Different levels of access for evaluators
Anthropic's blog post further highlighted how it is developing three tiers of access for evaluators. Each team in this tiered system would publish its own analysis of the company's risk posture and key decisions.
In the first tier, evaluators would conduct narrowly scoped risk investigations, with Anthropic deciding what kind of access they receive.
In the second tier, evaluators can publish their findings without Anthropic's approval and have access to more information. Anthropic can still require the removal of specific categories of information, though evaluators can publish their views on whether those redactions are justified.
In the third tier, evaluators receive the same access as METR, with employee-level access and publication rights similar to those in the second tier.
Anthropic has taken a lead in establishing such a tiered system, having been among the first to highlight the need for third-party evaluation in a paper following concerns raised by its AI researcher Jacob Coxon about the rapid pace of AI development and insufficient safeguards to contain risks from AI agents.
Handpicked stories, in your inbox
Global stories. Indian perspective. Zero noise.
No Spam. Unsubscribe Any Time.
First Published:Oct 16, 2026, 00:51:05 IST