Anthropic brings Accenture into internal model evaluations

Anthropic announced an evaluation partnership with Accenture in which Faculty, Accenture's AI business, will place evaluators inside the model developer. The arrangement is intended to give an external team access to employees, training decisions and internal systems that are usually unavailable in conventional assessments.

The work will examine alignment and safeguards, including adversarial testing of model behaviour. Anthropic will fund the evaluations directly, and the relationship will be non-exclusive. Both organisations also outlined plans for substantial investment in evaluation capacity over five years.

Several governance details remain unsettled, including standards for access, public reporting and funding arrangements as the programme develops. Anthropic retains responsibility for its models and deployment decisions. The partnership's practical value will depend partly on whether evaluators can investigate consequential issues and communicate findings with sufficient independence.