Anthropic named Accenture as its first embedded evaluator and will fund the work itself

Independent review of a frontier model has meant handing the model to a third party for a fixed window and reading the report afterwards. Anthropic is buying something else. Under a partnership announced on September 18, 2026, Accenture staff will work inside Anthropic with access the company describes as comparable to an employee's.

What an embedded evaluator gets

Anthropic describes the access in three parts. An embedded evaluator can watch models take shape in training, follow the decisions that govern how those models are built and deployed, and speak directly to employees. From that position it can assess how the company operates, check whether it is keeping its safety commitments, report incidents and give the public an account of benefits and risks.

The work itself is named in the announcement. Faculty, the specialist AI business Accenture runs, will lead it, and the scope covers evaluating and red-teaming models, conducting alignment assessments and testing model safeguards. Anthropic and Accenture each expect to invest at least $1 billion in building capacity over the next five years. That figure is stated as an expectation, not as a contract value.

• The evaluator can watch models take shape in training, not only after they are finished.
• It can follow the decisions that govern how a model is built and deployed.
• It can speak directly to employees.
• It can report incidents and give the public an account of benefits and risks.

Why the choice of firm is the news

Anthropic did not start with a nonprofit safety lab. It says it is in dialogue with METR and other nonprofit evaluators about piloting elements of embedded evaluation using their own funding. The Accenture partnership is non-exclusive, more evaluators are promised in the coming weeks, and Accenture will work with other AI developers in similar roles.

The reason Anthropic gives is practical. Accenture deploys AI for businesses and governments across many industries, and Anthropic argues that this understanding of how enterprises use AI in practice should inform how models are evaluated. The step is presented as progress on a commitment from the chief executive's essay, “We Must Pace the Frontier”.

What is not settled

Anthropic lists the gaps itself. There are no standards for what information an embedded evaluator should have access to, or for how it should report what it finds. There is no settled system for funding independent evaluation either, which is why Anthropic is funding Accenture's work directly.

That last point is the one to hold on to. Anthropic argues that funding should eventually come from pooled or government sources, as it called for in its Advanced AI Framework in June. Until such a source exists, the company paying for the audit is the company being audited. Anthropic does not pretend otherwise, and says the evaluators do not reduce its accountability.