Anthropic Taps Accenture as First Embedded AI Safety Evaluator

Anthropic has selected Accenture as its first embedded evaluator, advancing CEO Dario Amodei's three-step proposal to slow the pace of AI development and allow safety safeguards to catch up. The partnership, announced Sept. 20, will leverage Accenture's Faculty AI business to evaluate and red-team models, conduct alignment assessments, and test model safeguards with employee-like access to Anthropic's systems.

Amodei published his proposal on Sept. 12, warning that AI's capacity for recursive self-improvement could outrun humanity's ability to understand and control these systems if development continues unchecked. The first step calls for independent evaluators with deep internal access, a commitment Anthropic had already made unilaterally. OpenAI CEO Sam Altman and SpaceX CEO Elon Musk responded positively to the proposal, though Nvidia CEO Jensen Huang pushed back, arguing such regulation was unnecessary.

Both companies have committed at least $1 billion each to the initiative over the next five years. Because no existing system exists for funding independent AI evaluation, Anthropic will directly fund Accenture's work in the near term, with long-term funding expected to come from pooled or government sources. The partnership is non-exclusive, and Anthropic said it expects to announce additional evaluator partnerships in the coming weeks.

"Embedded evaluation is an emerging area, and we look forward to partnering with Anthropic to help accelerate the development of embedded evaluators, which we see as an important part of the safety landscape going forward," said Julie Sweet, chair and CEO of Accenture. Details of how the embedded evaluation will operate remain under development, as the model itself is new and untested at scale.