Anthropic Brings In Accenture as Its First Embedded AI Safety Evaluator

Update, Oct. 25: On Oct. 15, Anthropic said it had begun giving a six-person team from the nonprofit METR employee-like access, with METR free to publish its risk assessments, and that it is also working with SecureBio and Accenture as evaluators and is in early talks with others, including Hugging Face.

Anthropic has picked Accenture as the first outside firm to evaluate its models from inside the company. Faculty, Accenture’s specialist AI business, will lead work that includes “evaluating and red-teaming models, conducting alignment assessments, and testing model safeguards,” Anthropic said in a Sept. 18 announcement. Anthropic and Accenture each expect to invest at least $1 billion in building capacity for this work over the next five years.

The move is a first step toward a commitment in CEO Dario Amodei’s essay “We Must Pace the Frontier” to embed evaluators within Anthropic. Unlike today’s external testers, embedded evaluators will have access comparable to an employee’s, letting them watch models take shape in training, follow decisions on how models are built and deployed, and talk directly with staff. “From this vantage point, embedded evaluators can assess how a company operates, verify that it is keeping its safety commitments, and identify blind spots,” Anthropic wrote. “They can also report incidents and give the public a more informed account of benefits and risks.”

A surprising first pick

The choice surprised many AI watchers, TechCrunch reported, since discussion of embedded evaluators had centered on safety research groups such as METR, Redwood Research and Apollo Research. Accenture shares rose 8% after hours. Anthropic pointed to Accenture’s experience deploying AI for large companies and government agencies, and TechCrunch noted that as a large public company predating the AI boom, it’s also functionally independent of Anthropic. Accenture acquired Faculty in January.

Plenty is unsettled. Anthropic acknowledged there are no standards yet for what embedded evaluators should see or how they should report findings, and no settled funding system. It wants pooled or government funding in the long run, but for now will pay Accenture directly. The partnership is non-exclusive on both sides, and Anthropic said it was talking with METR and other nonprofits about piloting embedded evaluation with their own funding.

Some critics see industry-run evaluation as a way to sidestep accountability, TechCrunch noted. Anthropic’s response: embedded evaluators “do not reduce our accountability, but help to make it more verifiable. The safety of our models remains our responsibility.”