Anthropic and Accenture Bring Outside Evaluators Inside

Anthropic and Accenture announced a partnership on September 18 that places external evaluators alongside Anthropic’s internal teams and safety partners. The unusual part is their proximity: instead of judging models only from the outside, the team is intended to follow how they are built and deployed.

What employee-like access changes

Led by Accenture’s specialist AI business Faculty, the evaluators are expected to receive access comparable to an employee’s. That could let them observe model training, trace build and deployment decisions, and speak directly with staff. Their work will include model evaluation and red-teaming, alignment assessments, and tests of model safeguards.

Anthropic says the evaluators may also examine operations and safety commitments, identify blind spots, report incidents, and offer the public a better-informed account of benefits and risks.

Inspection does not mean enforcement

The team is not a regulator, statutory auditor, or enforcement body. It can inspect and report, but no formal power to halt training or block a model release has been announced. Anthropic also says independent evaluation does not reduce its own responsibility for model safety.

Large commitments, unfinished rules

Anthropic and Accenture each expect to invest at least $1 billion over the next five years, bringing the aggregate expected commitment to at least $2 billion. Anthropic will directly fund Accenture’s work for now, while operating standards, reporting rules, and long-term funding remain unsettled.

The arrangement is non-exclusive. Anthropic says it is also speaking with METR and other nonprofit evaluators, with additional evaluators expected to be announced in the coming weeks.