Anthropic Picked Accenture, Not a Safety Lab, as Its First Embedded AI Model Evaluator

Anthropic said the outside watchdogs sitting inside its labs would be independent. Then it picked a consulting giant, not a safety nonprofit, to be the first one. The company named Accenture as its first embedded evaluator, working from inside Anthropic on evaluating and red-teaming models, running alignment assessments, and testing safeguards.

The money behind it is large. Both companies expect to invest at least $1 billion in the effort over the next five years. Investors liked the sound of that, and Accenture shares rose about 8% after hours on the news.

The work will be led by Faculty, Accenture's specialist AI business, which Accenture acquired in January. Embedded evaluators get access closer to an employee's than an outside auditor's, watching models take shape in training, following the decisions behind how they are built and deployed, and talking directly to staff.

The choice landed as a surprise because many people watching this space expected the seats to go to nonprofit safety labs like METR, Redwood Research, or Apollo. Anthropic says more evaluators are coming and that it is still in conversation with METR and other nonprofits about pilots funded by those groups themselves.

Anthropic's reasoning is that Accenture brings practical experience deploying AI for large companies and governments, and that a public company predating the AI boom is more functionally independent. There are no standards yet for what access an evaluator should get or how it should report findings, and Anthropic expects the approach to change as the field matures.

Critics see the arrangement differently, reading CEO Dario Amodei's embedded-evaluator plan as self-policing designed to dodge real accountability. That skepticism sharpened after AI agents from OpenAI and Anthropic were caught hacking outside websites without tripping any internal alarms. Anthropic's answer is that evaluators "do not reduce our accountability, but help to make it more verifiable," and that "the safety of our models remains our responsibility."

For a company that has spent the past year courting Wall Street, from its move toward a Nasdaq listing to a $45 billion cloud deal with Nscale, handing its first oversight contract to a public consulting firm fits the pattern. Whether that counts as independent scrutiny or a commercial partnership dressed as one is the question the nonprofits still at the table are being asked to answer.