Anthropic is advancing its external evaluator program. The company stated that Faculty, AI division of Accenture, will join internally to participate in model evaluation, red team testing, alignment assessment, and security mechanism reviews.
This arrangement stems from an idea previously proposed by Anthropic’s CEO, Dario Amodei. The company states that both parties expect to invest at least $1 billion over the next five years to advance this project.
Why was Accenture selected?
Anthropic said one of the reasons for choosing Accenture is its greater experience in deploying AI within large enterprises and government institutions. Compared to security-focused organizations centered on cutting-edge research, Accenture has a stronger background in enterprise services.
The market responded noticeably to this choice. After the announcement, Accenture’s stock rose 8% in after-hours trading. Previously, external attention had been more focused on whether AI safety organizations such as METR, Redwood Research, and Apollo Research would participate.
More evaluators will be announced shortly.
Anthropic also stated that more evaluators will be announced in the coming weeks. The company is also in discussions with nonprofit organizations such as METR on how to pilot portions of "embedded evaluation" using their own funding.
The company also acknowledged that access rights and communication standards for external evaluators have not yet been standardized. Anthropic believes that such evaluations will not diminish the company’s responsibility but will instead help make model safety more verifiable.