Anthropic picked Accenture to audit it — and will pay the bill
Anthropic named its first embedded evaluator on Friday — a consulting firm the lab is paying directly, hours after outside experts published the conditions that arrangement would have to meet.
Anthropic says Accenture staff will work inside the company with access "comparable to an employee's," evaluating and red-teaming its models, running alignment assessments and testing safeguards. The partnership is led by Faculty, Accenture's specialist AI business, and Anthropic's own description of the access is the notable part: embedded evaluators watch models take shape in training, follow the decisions that govern how they are built and deployed, and speak directly to employees, rather than reviewing a finished model from the outside. That is roughly the vantage point the lab's own safety teams have, which is what Amodei proposed and what Amodei's pacing plan puts outside auditors inside Anthropic described when he published it last Saturday.
The funding is where it gets uncomfortable. Anthropic and Accenture each expect to invest at least $1 billion over the next five years in "building capacity in this area," but Anthropic will pay Accenture's bill itself. The lab acknowledged the contradiction: "Long-term, we think funding should come from pooled or government sources, as we called for in our Advanced AI Framework in June. As neither exists today, we plan to work with different evaluators under different funding arrangements." Anthropic framed the choice as a stopgap rather than the model — an auditor funded by the audited company until someone builds the public alternative.
Choosing Accenture is also a statement about what independence means. Faculty is not a frontier research lab, and Anthropic leaned on that: Accenture's experience deploying AI for large corporations and governments informs its safety read, and as a big public company that predates the AI boom, it is less entangled with the lab's own ecosystem than most research nonprofits working in AI safety. TechCrunch noted the lab also said no standards yet exist for evaluator access or communications, and that it expects the approach to evolve. The arrangement is non-exclusive; Anthropic says it is in talks with METR and others, with more evaluators to be announced in the coming weeks.
Critics read the same facts as the problem rather than the fix. Gary Marcus argued on X that "it ain't independent if you are partnering with them and chose the partner," and the incentive critique is sharper than the cultural one: METR is a nonprofit whose mission is transparency even when it costs it, while Accenture is a vendor whose client is the company being audited. Anthropic's answer is that embedded evaluators "do not reduce our accountability, but help to make it more verifiable." That is a testable claim, and it collides directly with the letter more than 100 researchers and evaluators published Friday morning, which demanded that evaluators take no payment contingent on findings, hold no other significant commercial business with the lab, and be shielded from retaliatory litigation — Over 100 AI experts attach conditions to the labs' evaluator pledge.
What to watch: whether Faculty's findings are ever published in a form Anthropic dislikes, and whether METR signs on with outside funding — that would show the lab is willing to be audited by someone it did not pick and does not pay.
If the lab chooses and pays the evaluator, what evidence would actually convince you the audit is real? Tell us in the comments.
