Anthropic and Accenture each expect to invest $1B in embedded AI evaluation

Anthropic and Accenture announced a partnership on Sept. 18, 2026, for independent evaluation of Anthropic’s frontier models from inside the company. Faculty, Accenture’s specialist AI business, will lead the work.

Both companies said they each expect to invest at least $1 billion over the next five years building capacity for the effort — expectations rather than a disclosed contractual cash commitment. Combined, that is about $2 billion in stated five-year capacity spending if both floors are met.

Anthropic said embedded evaluators will work alongside its internal teams and safety partners to evaluate and red-team models, conduct alignment assessments, and test safeguards. In Anthropic’s account, those evaluators will have access comparable to an employee’s, including during training and decisions about how models are built and deployed, and will be able to speak directly to employees. Accenture confirmed Faculty’s role and the evaluation tasks but did not describe that access in the same terms.

Anthropic said it will fund Accenture’s evaluation work directly, arguing pooled or government funding for independent evaluators does not yet exist. Accenture’s announcement did not describe how the work would be paid for.

Anthropic said the partnership is non-exclusive: it plans to name other evaluators in coming weeks, and Accenture will work with other AI developers in similar roles. Anthropic also said it is in dialogue with METR and other nonprofit evaluators about piloting elements of embedded evaluation with those groups' own funding.

CNBC framed the deal as Anthropic’s first named embedded evaluator and a concrete step toward CEO Dario Amodei’s recent “We Must Pace the Frontier” essay. TechCrunch noted Accenture acquired Faculty in January 2026. Secondary reports also place the evaluation work against a broader Accenture-Anthropic commercial relationship, including Claude training for Accenture staff.

This is paid, faculty-led evaluation with frontier-lab-scale access claims — not a safety essay. The open questions are independence (Anthropic is paying its evaluator while already a major commercial partner), what standards govern access and publication, and how the next evaluators are chosen and funded.