Anthropic Picks Accenture for AI Safety Testing

Anthropic appoints Accenture to conduct embedded third party safety evaluations of its AI models, according to the Financial Times. The five year agreement will be led by Faculty, the AI company that Accenture acquired in January.

Accenture teams will evaluate Claude models, try to break their safeguards and review Anthropic's safety procedures and alignment research. Anthropic will fund Accenture's work directly, while each company expects to invest at least $1 billion over the partnership.

Anthropic says the arrangement is nonexclusive and that future funding for independent evaluators should come from pooled or government sources. The company is also discussing separately funded evaluation work with the nonprofit Model Evaluation and Threat Research.