Anthropic has announced a collaboration with Accenture, integrating the consulting firm's AI division, Faculty, directly into its operations to conduct safety evaluations of its AI models. This initiative involves Accenture staff working inside Anthropic to perform tasks such as red-teaming, alignment assessments, and testing model safeguards. Both companies plan to invest at least $1 billion in this effort over the next five years.
The decision to partner with Accenture surprised many in the AI community, where expectations leaned toward specialized AI safety research organizations like METR, Redwood Research, or Apollo Research. Anthropic, which prioritizes AI safety and alignment, indicated that additional evaluators will be announced soon and that discussions are ongoing with nonprofit groups to pilot embedded evaluation funded independently.
Anthropic highlighted Accenture's extensive experience deploying AI solutions for large enterprises and government agencies as a key reason for the partnership. Unlike newer AI-focused entities, Accenture's established independence as a large public company provides a functional separation from Anthropic and the broader AI ecosystem.
The company acknowledged that standards for evaluator access and communication are still evolving and emphasized that external evaluations are becoming increasingly critical, especially following incidents where AI agents from OpenAI and Anthropic accessed external websites without internal alerts.
While some critics view this embedded evaluation approach as a potential means for self-regulation that might limit accountability, Anthropic maintains that these evaluators enhance transparency and verifiability without diminishing the company's responsibility for its models' safety.