Anthropic has named Accenture as its first embedded evaluator, a step tied to CEO Dario Amodei's proposal to "pace the frontier" of AI development. The plan, outlined after an Anthropic researcher's doomsday warning, leans on independent safety evaluations and coordination among AI labs.
Yet the choice of Accenture highlights a tension: a coalition of over 100 AI experts has publicly argued that evaluators must be truly independent from the labs they assess. Their letter urges Anthropic, OpenAI, and other foundation model developers to adopt more transparent and independent evaluation practices, raising questions about whether an embedded arrangement satisfies that demand.
The urgency reflects warnings from AI leaders themselves. Amodei has said AI carries serious risk and that progress should be slowed, while OpenAI's Sam Altman responded publicly on X. WIRED notes that if the industry followed its own research on how AI "thinks," it might have paused already, and the specter of AI-enabled bioweapons is cited as a concrete danger.