Anthropic tabs Accenture as embedded evaluator to help with AI slowdown proposal

Anthropic has named Accenture and its AI unit Faculty as its first "embedded evaluator" to help implement CEO Dario Amodei's Sept. 12 proposal to slow AI development and allow safeguards. The two companies plan to work on red-teaming, alignment assessments and testing of model safeguards, with each expecting to commit at least $1 billion over the next five years.

By AI Newsroom· Reviewed by Pranav, Founder & Editor-in-ChiefPublished 17 minutes agoUpdated 17 minutes ago0 views

Anthropic tabs Accenture as embedded evaluator to help with AI slowdown proposal

Why It Matters

The move seeks to create independent, hands-on scrutiny of advanced AI systems — a central element of Amodei's three-step plan to reduce the pace of development amid concerns over recursive self-improvement and catastrophic risks. Establishing such evaluators could shape how safety oversight is performed and funded as powerful models proliferate.

Key Facts

Proposal date: Sept. 12
First embedded evaluator: Accenture (and its AI business Faculty)
Planned investment: Each party expects to invest at least $1 billion over five years
Scope of work: Evaluating and red-teaming models, conducting alignment assessments, testing model safeguards
Funding arrangement: Anthropic will fund Accenture's work directly due to an absence of an existing funding system

Anthropic announced that it has selected Accenture, alongside Accenture’s AI unit Faculty, as the first embedded evaluator to support its plan to slow the pace of AI development. The partnership will focus on hands-on evaluation activities including red-teaming, alignment assessments and testing of model safeguards, part of CEO Dario Amodei’s three-step proposal published Sept. 12. The companies said details of how embedded evaluation will operate are still being worked out since the model is new. Anthropic and Accenture each expect to put at least $1 billion toward the effort over the next five years. Because no existing mechanism currently funds independent evaluation work long-term, Anthropic said it will bankroll Accenture’s efforts directly in the near term to address the urgency of the task. Amodei framed the proposal around concerns that rapid advances in AI, especially systems that can help build improved AI (a dynamic he referred to as recursive self-improvement), could outpace society’s ability to understand and control them. His call for embedded, independent evaluators drew positive reactions from some industry leaders, including OpenAI CEO Sam Altman and SpaceX CEO Elon Musk, while others such as Nvidia CEO Jensen Huang expressed disagreement with the need for such regulation. The partnership is non-exclusive, and Anthropic indicated it will announce additional evaluators in the coming weeks to expand the program. Accenture said its Faculty business already works on testing and evaluating models for leading AI labs and on building systems designed with safety and ethics in mind, and described embedded evaluation as an emerging component of the broader AI safety landscape.