Anthropic has partnered with consulting giant Accenture to provide independent, embedded evaluations of its frontier AI models for safety compliance. The move marks the first concrete step in a broader plan proposed last week by Anthropic chief executive Dario Amodei to slow the pace of AI development by requiring frontier AI companies to grant ongoing, employee-like access to third-party safety evaluators.

Under the arrangement, Accenture staff will be embedded within Anthropic, observing how models are trained, reviewing the decisions governing their development and deployment, and speaking directly with employees, with both companies committing to invest at least $1 billion each over five years. Anthropic said Accenture's experience helping businesses and governments deploy AI gives it useful insight into enterprise use of its models, though it acknowledged that the embedded-evaluator concept is new, with no established standards yet for scope, access or reporting procedures. Anthropic added that it plans to work with further evaluators beyond Accenture, with further announcements expected in the coming weeks, and noted that Accenture may also work with other AI firms.