Anthropic gives Accenture staff-level access to evaluate its own models
Embedded evaluators get more than an outside auditor's usual view, including direct interaction with staff and visibility into deployment decisions, in a partnership both companies say they'll fund with at least $1 billion each over five years.
Anthropic announced a partnership on September 18 with Accenture's AI unit, Faculty, to embed evaluators inside Anthropic's operations. They will run model evaluation, red-teaming, alignment assessments and safeguard testing with, in Anthropic's description, "access comparable to an employee's." Accenture's evaluators can observe model development as it happens, follow deployment decisions and talk directly with Anthropic staff, instead of reviewing finished artifacts after the fact.
Who pays
Both companies say they expect to invest at least $1 billion each in building this capacity over the next five years. Anthropic is funding Accenture's work directly for now, but says that shouldn't be permanent: "long-term, we think funding should come from pooled or government sources." A lab paying its own evaluator, however embedded and well-resourced, is not the end state. Anthropic also calls Accenture the first of several and says it is in discussions with METR and other nonprofit evaluators about similar embedded work.
Access and its limits
An evaluator inside the building, talking to the team that shipped the model, can catch things a document review can't. That is a step beyond the external audits and published safety cards labs have relied on so far. It is still Anthropic choosing, funding and granting access to its own evaluator, which is the dependency the "pooled or government sources" line tries to get ahead of. If you're deciding how much to trust a vendor's safety claims, count this as more rigorous than self-attestation, under a funding structure Anthropic itself says isn't final.