Anthropic brought in Accenture to place independent evaluators inside the company, the first concrete move on the pledge Dario Amodei made in his September essay urging the industry to slow down. Faculty, Accenture’s AI arm, will run the work, which covers red-teaming, alignment assessments, and testing of model safeguards, and each side expects to put at least $1B into it over five years. These evaluators get access on par with an employee’s, letting them watch models during training, follow the decisions behind how each one is built and shipped, question staff, and report incidents publicly. Anthropic is paying Accenture directly, which it says isn’t the long-term answer, since it wants evaluation funded from pooled or government money that doesn’t exist yet.
