Partnering with Accenture on embedded evaluation
anthropic ·
Anthropic is partnering with Accenture’s specialist AI business, Faculty, to conduct independent evaluations of frontier AI models. The work will include model evaluation and red-teaming, alignment assessments, and testing safeguards, drawing on Accenture’s experience with enterprise and government AI deployments. Anthropic and Accenture each expect to invest at least $1 billion in building evaluation capacity over the next five years. Embedded evaluators will work inside AI companies with employee-like access, allowing them to observe training, examine development and deployment decisions, speak with staff, verify safety commitments, identify blind spots, and report incidents. Anthropic says independent evaluators will strengthen the verifiability of its accountability, while responsibility for model safety will remain with Anthropic. Because standards and long-term funding systems have not yet been established, Anthropic will initially fund Accenture’s work directly while also speaking with METR and other nonprofit evaluators, and it expects the non-exclusive approach to evolve toward a broader ecosystem with shared standards.
anthropic