Anthropic just hired an outside auditor for its AI models - and is paying that auditor's bill itself.
Anthropic named Accenture as the first outside firm embedded to evaluate its frontier models before and after release. The deal sits inside a broader partnership in which each company expects to invest at least $1 billion over five years, a combined $2 billion commitment to build out Claude deployments across Accenture's client base. Anthropic is funding Accenture's specific evaluation work itself, drawing on its own side of that investment. In the same announcement, Anthropic said evaluation funding should ultimately come from pooled industry funds or government money - because no such independent pool currently exists.
An evaluator paid by the company it's grading is a textbook conflict of interest, and Anthropic is naming that problem out loud rather than pretending it doesn't exist. That's more candor than most AI labs offer, but candor doesn't fix incentives: Accenture now has a commercial relationship with Anthropic worth billions, on top of the fee for judging Anthropic's models.
Call it evaluation theater with better disclosure - useful until someone actually builds the independent funding pool Anthropic says should exist.