Anthropic said Friday, September 18, 2026, that it is partnering with Accenture on independent evaluation of frontier AI — a concrete step toward the “embed evaluators” commitment in CEO Dario Amodei’s essay “We Must Pace the Frontier.”
The work will be led by Faculty, Accenture’s specialist AI business. Scope covers evaluating and red-teaming models, running alignment assessments, and testing model safeguards. Anthropic’s company post says Accenture’s enterprise-deployment experience will inform how those checks are designed.
Anthropic and Accenture each expect to invest at least $1 billion over the next five years in building capacity for the work. Reuters frames the pair of commitments as roughly $2 billion combined. Bloomberg also covered the embed-evaluators announcement.
What “embedded” means
Unlike today’s external auditors, Anthropic says embedded evaluators will work inside AI companies with access comparable to employees: watch models take shape in training, follow build and deploy decisions, speak directly to staff, verify safety commitments, surface blind spots, and report incidents publicly.
Anthropic stresses that independent embedded evaluators do not reduce the lab’s accountability — they make it more verifiable. Model safety remains Anthropic’s responsibility.
Funding and a non-exclusive ecosystem
Standards for what embedded evaluators can see, and how they report, do not exist yet, Anthropic said. Long-term, the company wants pooled or government funding. For now, Anthropic will fund Accenture’s work directly. It is also in dialogue with METR and other nonprofit evaluators to pilot pieces of embedded evaluation under different funding.
The partnership is non-exclusive. Anthropic says more evaluators are coming in the coming weeks, and Accenture will work with other AI developers in similar roles. The pitch is an ecosystem of evaluators with shared standards, not a single contracted auditor.
This brief covers Anthropic’s Friday partnership announcement and the major-outlet corroboration. It is distinct from Anthropic’s R&D Automation Index explainer and from Hill / California interrupt and kill-switch policy coverage.