Independent Evaluators Now Work Inside Anthropic
On Sep 18, 2026, Anthropic announced a partnership with Accenture's Faculty AI unit to embed independent evaluators inside the company. Unlike external auditors, embedded evaluators get employee-level access—watching models during training, tracking deployment decisions, and verifying safety commitments from the inside. Anthropic and Accenture each plan to invest at least $1 billion over five years.
Starting Sep 18, 2026, independent evaluators work inside Anthropic—not alongside it. That distinction matters. External auditors review AI systems after the fact. Embedded evaluators watch models take shape during training, follow the decisions about how those models are built and deployed, and speak directly to employees. From that position, they can verify that safety commitments are being kept and report what they find to the public.
The first partner is Faculty, Accenture's specialist AI unit. Faculty's work will cover evaluating and red-teaming Claude models, conducting alignment assessments, and testing model safeguards. Accenture helps enterprises and governments deploy AI across many industries—that real-world deployment experience shapes how they approach safety evaluation. Both organizations plan to invest at least $1 billion each over the next five years building capacity in this area.
Embedded evaluation is new, and the standards don't exist yet. There are currently no agreed rules for what information embedded evaluators should access, how they should report findings, or who should fund the work. Anthropic is funding Accenture's work directly for now, while longer-term pooled or government funding models are developed—as called for in Anthropic's Advanced AI Framework in June. METR and other nonprofit evaluators are also being piloted in parallel under their own funding. The partnership is non-exclusive: Anthropic plans to work with additional evaluators, and Accenture may work with other AI developers.
For enterprise teams using Claude, this is the accountability structure made checkable. Anthropic can state that Claude is safe; now an independent organization with employee-level access can verify that claim. The goal is an ecosystem of evaluators working with shared standards—not a single audit that passes or fails a model, but ongoing oversight embedded in how frontier AI is built.
Key takeaways: Embedded evaluators get employee-level access—they watch training, follow decisions, and speak to staff. Faculty (Accenture's AI unit) will cover red-teaming, alignment assessments, and safeguard testing. Both organizations plan at least $1 billion each over five years. No standards for embedded evaluation exist yet; Anthropic is funding Accenture directly while longer-term models develop. The partnership is non-exclusive, and more evaluators will be announced.