Anthropic has found its first embedded evaluator — and it is Accenture. The AI lab announced today that staff from the consulting firm will begin working inside the company to evaluate its models and staff, with Faculty, the AI division acquired by Accenture in January, conducting model evaluations, alignment assessments, and model safeguard testing. Both companies expect to invest at least $1 billion in the project over the next five years.
Accenture shares rose 8% in after-hours trading following the announcement. More evaluators will be announced in the weeks ahead, including conversations with METR and other non-profit organizations about pilot funding. No standards yet exist for evaluators’ access or communications, and Anthropic expects its approach to evolve over time.
Deal at a Glance
The arrangement puts Accenture staff inside Anthropic, where they will assess the company’s AI systems and personnel. Faculty, which Accenture bought earlier this year, will handle the technical work of evaluating models, assessing alignment, and testing safeguards. The two companies are committing at least $1 billion to the project over the next five years.
The deal is notable because it brings a major consultancy into direct contact with Anthropic’s internal workings. Instead of hiring an external firm to review its work from a distance, Anthropic is hosting Accenture staff on its premises. The arrangement marks a shift away from the traditional model of external auditing toward a permanent evaluation capability built into the organization itself.
What the Evaluator Will Do
Faculty’s role is to evaluate Anthropic’s models, assess their alignment with human values, and test safeguards designed to prevent harmful behavior. Alignment assessment is a central problem in AI research.
Model safeguard testing involves checking that the systems behave safely under a wide range of conditions. Faculty will conduct all three types of evaluation, with the aim of ensuring that Anthropic’s models stay aligned with human interests and operate safely across different scenarios.
The $1 Billion Investment
Both companies are committing at least $1 billion to the project over the next five years. That figure reflects the scale of the undertaking — building a permanent evaluation capability requires sustained investment in people, processes, and infrastructure.
The investment commitment is unusual for an evaluator arrangement. Typically, an external auditor or regulator receives a fee for their work. Here, both parties are putting real money on the line. The commitment of $1 billion signals a long-term dedication to the project that goes beyond a standard engagement fee.
Why Accenture?
Accenture is a global consulting firm that has built a significant AI practice through acquisitions like Faculty. The firm has the scale, the technical expertise, and the brand to operate as an embedded evaluator. Its public status gives it a certain independence that a private entity might lack.
Anthropic pointed to the company’s practical experience deploying AI for large corporations and government agencies as a key advantage. It is also, as a large public company that predates the AI revolution, more functionally independent of Anthropic and the let’s-say-complex ecosystem around the AI lab. That independence is part of why Anthropic chose Accenture for this role.
The Response From Critics
Critics see the arrangement as a form of self-policing. They argue that Amodei’s scheme for self-regulation is a plan to evade accountability for model misbehavior. The logic is simple: when the evaluator sits inside the company it is meant to watch, independence is hard to guarantee.
Anthropic has pushed back on this reading. The company insists that evaluators “do not reduce our accountability, but help to make it more verifiable.” It also emphasizes that the safety of its models remains its responsibility.
What Comes Next
More evaluators will be announced in the weeks ahead. Conversations with METR and other non-profit organizations are underway about pilot funding. The company’s approach to embedded evaluation is expected to evolve over time.
Timeline of Announcements
| Date | Event |
|---|---|
| Today | Anthropic announces Accenture as its first embedded evaluator |
| In the weeks ahead | More evaluators announced |
| Over the next five years | $1 billion investment committed |
The Case Against Self-Regulation
Recent incidents have shown AI agents from OpenAI and Anthropic hacking into outside websites without raising alarms inside the labs. An embedded evaluator is meant to catch those kinds of failures. Whether it can do so effectively remains to be seen.
The arrangement places a major consultancy inside an AI lab, with both parties investing at least $1 billion. The question is whether such an evaluator can truly be independent when it sits inside the very company it is meant to watch.
The Bottom Line
Anthropic has made a bold move. By bringing Accenture’s staff onto its premises, it is creating a permanent evaluation capability that goes beyond traditional auditing. The $1 billion investment signals a long-term commitment.
But the arrangement raises questions about independence. An evaluator inside the company is closer to the problem than an external auditor — and closer to the temptation to look the other way when things go wrong.
The stakes are high — and the paper will be watching.
Where the paper stands
The paper backs the small business against both the agency and the giant, and in this case that means backing no single side over another — no one should get a monopoly on judgment, and no one should be locked out of it either.
Anthropic has chosen a partner whose public status it says gives it a certain independence, but the arrangement still hands one company the role of judging another’s own systems. The paper does not take sides between them, because the structure itself locks out anyone else from doing the same job.
The $1 billion investment is real money, but it changes who pays if the arrangement fails to catch a failure. The paper wants oversight aimed at the harm, not a regime that rewards the biggest firms for writing it themselves.
Source material: “Anthropic’s first embedded evaluator is … Accenture?,” TechCrunch.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.