Anthropic and Accenture to Invest $2 Billion in AI Evaluation Partnership

Anthropic and Accenture have announced a significant partnership aimed at enhancing the evaluation of frontier AI models, committing at least $1 billion each over the next five years. This collaboration will see Accenture embed evaluators within Anthropic, allowing for in-depth scrutiny of AI safety practices.

The initiative is part of Anthropic's broader strategy to allow independent evaluators permanent access to its AI models, particularly the Claude family. By providing evaluators with employee-level access, the partnership is designed to bolster transparency and accountability in AI development.

Led by Faculty, Accenture’s specialist AI business, the evaluators will work closely with Anthropic teams to conduct alignment assessments, red team models, and test model safeguards. This embedded approach is expected to provide evaluators with insights comparable to those of Anthropic employees, enabling them to observe training processes and assess safety commitments directly.

While the framework for this partnership remains experimental, Anthropic has previously advocated for independent evaluations to be financed through pooled or government-backed mechanisms. For this specific collaboration, however, Anthropic will directly fund the evaluation efforts.

Additionally, Anthropic is in discussions with nonprofit evaluator METR and plans to announce more evaluation partners soon. The arrangement allows Accenture to collaborate with other AI developers, emphasizing that the responsibility for model safety remains with Anthropic.