Anthropic and Accenture Each Pledge $1B for Embedded AI Evaluators

Anthropic formally announced on September 18 that Faculty, Accenture's specialist AI business unit, will serve as its first embedded evaluator, with both Anthropic and Accenture each committing at least $1 billion over five years. Faculty staff receive red-teaming and alignment-testing access comparable to an Anthropic employee. The arrangement is non-exclusive, and Anthropic is also in talks with research nonprofit METR to pilot a similar embedded evaluation program using METR's own funding. Anthropic is directly funding Faculty's specific evaluation work, with a stated expectation that long-term funding should come from pooled or government sources.

The partnership implements the first stage of a plan Dario Amodei outlined in his September 12 essay 'We Must Pace the Frontier,' in which Anthropic unilaterally committed to giving external evaluators ongoing, employee-level access to verify safety practices, report incidents, and assess training pipelines, not just completed models. Amodei named METR as an example partner and drew on banking industry precedent. The essay's second stage calls for democratic coordination among frontier labs requiring a narrow antitrust waiver from Washington, and the third for global coordination including chip export controls and protecting model weights against distillation attacks.

Amodei stated that recursive self-improvement has been underway at Anthropic and across the industry since roughly this summer, with internal models Mythos and Astra initiating early-stage RSI. He warned that within 6-12 months an AI-enabled botnet swarm could cause hundreds of billions of dollars in damage.