Anthropic plans to bring outside safety evaluators into its development work. On September 18, 2026, it announced a partnership with Accenture, whose AI business Faculty will lead the team. The proposed work covers model testing, alignment assessments and checks on safeguards.
The evaluators are meant to receive access comparable to employees. That would let them observe training and speak with staff rather than test only a finished model. Anthropic presents the arrangement as a step toward Dario Amodei's proposal for external oversight.
Each company expects to invest at least $1 billion in this area over five years. Those are planned investments, not money already spent. Anthropic will initially pay for Accenture's evaluation work itself. The arrangement is non-exclusive, with further evaluators expected to join.
Independence therefore needs clear rules. Anthropic acknowledges that standards for access, reporting and funding remain unsettled. It says discussions with METR include pilots under different funding arrangements. Meanwhile, the company intends to keep training and releasing models. The announcement establishes neither a development pause nor a demonstrated improvement in safety.