Anthropic and Accenture Bet $2B on Embedded AI Safety Audits

Anthropic is partnering with Accenture's Faculty unit for embedded, ongoing evaluation of frontier models, according to a report published by alphasignal.ai. Alphasignal.ai describes the evaluation work as embedded and ongoing rather than a single review, and lists the scope as covering frontier models.

The stated scope covers model evaluation, red-teaming, alignment assessments and safeguard testing. The arrangement includes employee-level access, according to the report. Alphasignal.ai describes the agreement as the first disclosed implementation of a proposal by Anthropic CEO Dario Amodei to place outside evaluators inside frontier AI labs. The report identifies Accenture's Faculty unit as the counterparty, with the two companies as the named parties to the arrangement.

The report does not state which models fall within scope, what employee-level access involves in practice, or how any findings would be published or shared. It gives the five-year period as the timeframe for the two $1 billion commitments, with each company committing at least that amount. The report's framing emphasises that the evaluators are placed inside the lab rather than reviewing it from outside, and that the work is ongoing rather than a one-time assessment. The report presents the partnership as the first publicly disclosed instance of the outside-evaluator proposal being put into practice, and attributes that proposal to Amodei. The report does not describe any regulatory requirement underlying the arrangement.