Anthropic gives outside AI evaluators employee-level access to its systems
US artificial intelligence company Anthropic has onboarded six employees of the independent research group METR and given them access to internal tools. The firm is trying to reassure investors that safety concerns will not overshadow its expected stock market flotation.
Anthropic, the AI developer, has given six staff from METR, an independent group that evaluates risks posed by AI models, access roughly comparable to that of its own employees. The arrangement was reported by the Financial Times.
The company says the METR evaluators have received company hardware, office space and access to internal tools and systems roughly on par with what its own risk assessment teams can use. Anthropic has encouraged its staff to share information with the evaluators as they would with a fellow employee working on risk. It compared the approach to banking: when a bank's risk-taking could threaten the public, third parties try to see that in advance.
Why it matters now
Anthropic also faces pressure to reassure investors that safety issues will not overshadow its initial public offering, expected this year at a valuation above $2tn. It devoted a large part of this week's investor presentations to its safety practices. METR was among the organisations that investigated the hack of Hugging Face by rogue OpenAI models, reviewing thousands of logs and receiving company laptops.
Rules and funding
Anthropic is not paying METR for the arrangement. The research group does not accept donations from AI companies or their staff, and is funded by individuals, philanthropic organisations and government bodies such as the UK's AI Security Institute. The evaluators will publish their analysis and Anthropic cannot veto it. It can, however, require the redaction of certain categories of information, including customer data, material covered by attorney-client privilege and details of model architectures.
Anthropic is also still working with Accenture, the consultancy named a safety-testing partner last month; each has committed at least $1bn to set up new safeguards. Another outside tester, SecureBio, reviews the company's chemical and biological weapons risks.
"We're embedding evaluators voluntarily, and we believe every frontier AI company should do the same," Anthropic said. METR responded: "We stand ready to work with all frontier AI developers to bring better evidence to the public."
