Anthropic and Accenture to Invest $1 Billion Each Over 5 Years to Introduce Embedded AI Model Evaluations

Anthropic has signed an embedded evaluation partnership with Faculty, Accenture's AI-focused division. This is part of the commitment outlined in the CEO's essay "We Must Pace the Frontier," representing a new safety approach where independent evaluators verify models from within the AI company.

The core of the partnership is for evaluators to gain access rights similar to employees, directly observing the model training process and deployment decisions. This allows them to verify compliance with safety commitments, identify blind spots, and report incidents. Accenture will conduct model alignment evaluations and red-teaming, leveraging its experience with AI adoption in enterprises and government agencies.

Both companies plan to invest at least $1 billion each over the next five years to build capabilities in this area. Although there are currently no standards or funding systems for embedded evaluations, Anthropic is directly supporting Accenture's work and collaborating with non-profit evaluation organizations such as METR to conduct pilots. This partnership is non-exclusive, and Anthropic plans to announce collaborations with other evaluation organizations in the coming weeks.