Partnering with Accenture on embedded evaluation

What it claims

Anthropic has started an embedded evaluation partnership with Accenture, run by Accenture's AI business Faculty. The partnership is non-exclusive on both sides. Anthropic says it will announce other evaluators "in the coming weeks," and Accenture will do similar work for other AI developers. Anthropic presents the move as a step toward a commitment in its CEO's essay "We Must Pace the Frontier" to embed evaluators inside the company.

How it works

Embedded evaluators work inside the company with access comparable to an employee's. The page lists what that access covers:

• watching models take shape during training
• following the decisions that govern how models are built and deployed
• speaking directly to employees

From that position, the page says evaluators can do several things. They can assess how the company operates, verify that it keeps its safety commitments, and identify blind spots. They can also report incidents and give the public an account of benefits and risks. Anthropic states that this does not reduce its own accountability: the safety of its models remains Anthropic's responsibility. The aim is to make that accountability verifiable.

The page describes three funding arrangements:

• Accenture: Anthropic funds the work directly, citing urgency.
• METR and other nonprofits: these evaluators would pilot parts of embedded evaluation with their own funding. Talks are ongoing.
• Long term: Anthropic wants funding from pooled or government sources, as it proposed in its Advanced AI Framework in June. Neither source exists yet.

Limits

The page says many operating details are still being worked out. No standards exist for what information embedded evaluators should see or how they should report findings. There is also no settled funding system.

The page names no reporting channel, publication schedule, disclosure rights or veto power for evaluators. It also does not say which models are in scope. Because Anthropic pays Accenture directly, readers judging how independent the arrangement is have only Anthropic's statement to go on. Anthropic says its approach will change as more evaluators join.