Anthropic said on Friday that it selected Accenture as its first embedded evaluator, with staff from Faculty, Accenture’s AI division, set to work inside the company to red-team models, test safeguards and assess alignment. CNBC reported the announcement the same day, and TechCrunch said Anthropic’s blog post described Faculty’s remit in those terms.
Anthropic and Accenture said they expect to invest at least $1 billion over five years in the effort. CNBC reported that Anthropic said it will fund Accenture’s work directly for now because pooled or government funding does not yet exist; TechCrunch reported that Anthropic pointed to Accenture’s experience deploying AI for large corporations and government agencies as part of its rationale.
Anthropic said the arrangement is non-exclusive, that it is also in discussion with METR and other nonprofits, and that no standards yet exist for evaluator access or communications. The company said more evaluators will be announced in the coming weeks.
For organizations assessing evaluator claims, the useful next check is whether the evaluator’s scope, system access, publication rights and route for escalating findings are independently defined. Those details show whether “embedded” means useful scrutiny, credible independence or both.
