Anthropic and Accenture put 2 billion dollars into embedded AI evaluation

Anthropic has named Accenture as an embedded evaluator for its frontier AI models, turning Dario Amodei’s call for inside-the-lab safety oversight into a concrete commercial partnership.

The work will be led by Faculty, Accenture’s specialist AI business. Anthropic says the team will evaluate and red-team models, conduct alignment assessments and test safeguards with access closer to an employee than a normal outside reviewer. The idea is to watch model development and deployment decisions while they are happening, not only audit a finished system from the outside.

The financial commitment is unusually large for AI safety infrastructure: Anthropic and Accenture each expect to invest at least 1 billion dollars over five years to build capacity for this work. Anthropic says the arrangement is non-exclusive and that it is also talking with METR and other nonprofit evaluators.

The choice of Accenture is notable because much of the public discussion around embedded evaluation has focused on safety research groups. Anthropic argues that Accenture brings practical experience from enterprise and government AI deployments. The unresolved question is independence: Anthropic says embedded evaluators do not reduce its accountability, but make safety commitments more verifiable.