Anthropic has selected Accenture as its first embedded evaluator, a move aimed at supporting CEO Dario Amodei's proposal to decelerate the pace of artificial intelligence development and allow time for safety safeguards to take shape.
Amodei laid out his three-step proposal on Sept. 12, citing warnings that the swift, unchecked advancement of AI carries the potential for catastrophic harm. The core concern centers on what he described as recursive self-improvement — the growing ability of AI systems to build subsequent generations of AI — which, left unchecked, he wrote "could outrun our ability to understand and control these systems."
The first step in the proposal calls for independent evaluators granted employee-like access to AI systems. Amodei said Anthropic had already unilaterally committed to that step before announcing the Accenture partnership on Friday. Under the arrangement, Accenture and its AI division, Faculty, will evaluate and red-team models, conduct alignment assessments, and test model safeguards.
Details on how the embedded-evaluation model will operate are still being finalized, Anthropic acknowledged, noting that embedded evaluation is a new area. Julie Sweet, chair and CEO of Accenture, called it "an emerging area" and said the firm looks forward to helping accelerate the development of embedded evaluators as an important part of the safety landscape.
Anthropic and Accenture each expect to invest at least $1 billion in the project over the next five years. Amodei wrote that no existing system funds independent evaluation and that long-term financing should come from pooled or government sources. Given the urgency, however, Anthropic will fund Accenture's work directly.
The partnership is non-exclusive. Anthropic expects to announce additional evaluators in the coming weeks.
Amodei's proposal drew mixed responses from industry leaders. OpenAI CEO Sam Altman and SpaceX CEO Elon Musk responded positively. Nvidia CEO Jensen Huang declined, arguing that such regulation was unnecessary.
