Anthropic's plan to bring outside safety evaluators inside its labs has its first participant: Accenture. Staff from the consulting giant will work within the company to scrutinize its models and employees, Anthropic said in a blog post.

The work falls to Faculty, the AI division Accenture acquired in January. Its tasks will include evaluating and red-teaming models, running alignment assessments, and testing model safeguards. The two companies expect to invest at least $1 billion in the project over five years.

The pick raised eyebrows among AI watchers and lifted Accenture's shares 8% after hours. Talk of embedded evaluators had centered on safety research groups like METR, Redwood Research, and Apollo Research - especially at Anthropic, which places safety and alignment at its core. Anthropic said more evaluators will be named in coming weeks and that it is talking with METR and other nonprofits about piloting embedded evaluation with their own funding.

Accenture is not a deep learning research powerhouse, but Anthropic cited its experience deploying AI for large corporations and government agencies, plus its independence as a public company. The lab acknowledged no standards yet govern evaluator access or communication, and said its approach will evolve. External evaluations are already standard in model releases, but recent incidents in which OpenAI and Anthropic agents hacked outside websites without internal alarms have raised the stakes.

Critics who want more responsible AI development call the self-policing scheme a way to dodge accountability. Anthropic maintains the evaluators do not reduce its accountability but make it more verifiable, adding that model safety remains its responsibility.

Source: TechCrunch