Partnering with Accenture on embedded evaluation

Anthropic has announced a partnership with Accenture for independent evaluation of frontier AI. The company describes it as a significant move toward a pledge in its chief executive's essay "We Must Pace the Frontier", namely to place evaluators inside Anthropic. Faculty, Accenture's specialist AI business, will lead the work, which covers evaluating and red-teaming models, running alignment assessments and testing safeguards. Anthropic and Accenture each expect to put in at least $1 billion over five years to build capacity in this area.

Anthropic explains that embedded evaluation is new and that many operational details remain undecided. Unlike current external evaluators, embedded ones would work inside AI companies with access similar to an employee's. That would let them watch models develop in training, follow the decisions behind how models are built and deployed, and talk directly to staff. From there they could judge how a company operates, check that it keeps its safety commitments, find blind spots, report incidents and give the public a better-informed account of benefits and risks. Anthropic stresses that this makes its accountability more verifiable rather than smaller, and that model safety stays its own responsibility.

The post acknowledges gaps. No standards yet govern what information embedded evaluators should see or how they should report, and there is no settled way to pay for independent evaluation. Anthropic thinks funding should eventually come from pooled or government sources, as its Advanced AI Framework in June urged. For now it will fund Accenture directly and is talking with METR and other nonprofits about piloting elements using their own money.

On status, the arrangement is non-exclusive: Anthropic says more evaluators will be announced soon, and Accenture is to serve other developers in comparable roles. Anthropic expects its approach to change as the field matures.