Anthropic’s latest answer to the AI-safety accountability problem is not another public pledge. It is an auditor with an employee-like view from inside the company.

The move follows chief executive Dario Amodei’s commitment to embed outside evaluators within Anthropic as its models become more powerful. The company has now selected Accenture, through its specialist AI business Faculty, as its first such partner. Accenture will red-team models, assess alignment and test safeguards, while gaining access comparable to that of an Anthropic employee.¹

Anthropic says that proximity matters. An embedded evaluator can watch models develop during training, trace the decisions behind their release and deployment, and speak directly to staff, it argues. From that position, evaluators could identify blind spots, report incidents and give the public a clearer account of AI’s risks and benefits. “Independent embedded evaluators do not reduce our accountability, but help to make it more verifiable,” Anthropic said.¹

The financial scale is striking: Anthropic and Accenture each expect to invest at least $1 billion over five years to build evaluation capacity.¹ Yet the company also concedes that the system is unfinished. There are no established standards governing evaluator access or reporting, and no settled funding model. For now, Anthropic will directly fund Accenture’s work, even as it says that pooled or government-backed funding would be preferable over time.¹

That arrangement is where the doubts land. One skeptical framing asked simply: “Anthropic’s first embedded evaluator is ... Accenture?”² The concern is not that evaluation is unnecessary, but that a consulting firm paid by the company it examines may struggle to look fully independent.

Anthropic insists the partnership is non-exclusive and says it is also talking with nonprofit evaluators including METR. Its bet is that scrutiny from within can be made credible. Its critics are asking whether the watchdog has been invited too far inside the gate.