Anthropic announced on September 18, 2026 that it is partnering with Accenture’s specialist AI business, Faculty, to pilot embedded, independent evaluation of frontier artificial intelligence systems.
What the partnership involves
The collaboration will place evaluators and red teams inside Anthropic to assess models, perform alignment evaluations, and test model safeguards. Unlike current external evaluators, these embedded evaluators will have access comparable to an employee’s, enabling them to observe model training, follow the governance decisions that shape model development and deployment, and speak directly with staff.
From that vantage point, embedded evaluators can review how the company operates, verify whether Anthropic is meeting its safety commitments, identify blind spots, report incidents, and provide the public with a more informed account of benefits and risks.
Anthropic framed this step as implementing a commitment made in its CEO’s essay “We Must Pace the Frontier.” The company emphasized that independent embedded evaluation is intended to make accountability more verifiable, not to reduce Anthropic’s responsibility for model safety.
Funding and capacity commitments
Anthropic and Accenture each expect to invest at least $1 billion over the next five years to build capacity for these evaluations. Because there is not yet a settled funding system for independent evaluation, Anthropic will directly fund Accenture’s initial work.
Anthropic is also in dialogue with METR and other nonprofit evaluators to pilot elements of embedded evaluation funded by those organizations. In its June Advanced AI Framework, Anthropic argued that long‑term funding for independent evaluation should come from pooled or government sources; such mechanisms currently do not exist.
Standards, independence, and non‑exclusivity
There are no established standards yet for what information embedded evaluators should access or how they should report findings. Anthropic and Accenture characterize the partnership as non‑exclusive: Anthropic intends to work with multiple evaluators in parallel and will announce additional partners in the coming weeks, while Accenture will provide similar services to other AI developers.
Why this matters
Embedded evaluation introduces a different model for assessing frontier AI safety by giving evaluators early and deep visibility into model development and deployment choices. The goal is to improve transparency and the verifiability of safety commitments while maintaining that ultimate responsibility for model safety remains with the developer. Anthropic says it will continue to train and release frontier models with independent evaluators working alongside them, and that its approach will evolve as the field matures.
