Anthropic partners with Accenture on embedded AI evaluators
In brief
- Anthropic selects Accenture as first embedded evaluator to slow AI development pace
- CEO Dario Amodei's September proposal aims to align safety safeguards with rapid AI advancement
- Partnership is non-exclusive; additional evaluators expected in coming weeks
The AI acceleration problem
AI has been advancing drastically faster, driven primarily by AI's growing ability to build the next generation of AI — a dynamic Amodei called recursive self-improvement. This acceleration has outpaced the industry's ability to develop adequate safeguards. Amodei's proposal sought to address this gap.
The first step in the plan involves independent evaluators with deep, employee-like access to AI systems. These evaluators can assess alignment, conduct red-teaming exercises, and test model safeguards from the inside. Anthropic needed partners to make this work.
Accenture partnership details
Anthropic announced the partnership with Accenture and its AI business Faculty on Friday. The collaboration will focus on evaluating and red-teaming models, conducting alignment assessments, and testing model safeguards. It's a significant commitment: Anthropic and Accenture each expect to invest at least $1 billion in the project over the next five years.
Anthropic will fund Accenture's work directly given the urgency of the effort, though the company noted there's no existing system for funding independent evaluation long-term. Future funding should come from pooled or government sources.
What's next
The partnership is non-exclusive, and Anthropic expects to announce other evaluators in forthcoming weeks. The company's September proposal drew mixed reactions from industry leaders. OpenAI CEO Sam Altman and SpaceX CEO Elon Musk responded positively, while Nvidia CEO Jensen Huang took a different view, arguing that such regulation wasn't necessary.
The Accenture deal signals Anthropic's intent to move forward regardless. Building a credible independent evaluation ecosystem takes time and capital. This partnership is a concrete step toward that goal.
Frequently asked questions
What is recursive self-improvement in AI?
Recursive self-improvement is the dynamic where AI systems use their own capabilities to build the next generation of AI, accelerating development beyond what humans alone can achieve. This acceleration has outpaced the industry's ability to develop safeguards.
What was Amodei's proposal to slow AI development?
CEO Dario Amodei published a three-step proposal on September 12 to slow AI development and allow safeguards to keep pace. The first step involves independent evaluators with employee-like access to AI systems for red-teaming, alignment assessments, and safety testing.
Why is Anthropic partnering with Accenture?
Anthropic chose Accenture to serve as its first embedded evaluator, implementing the independent evaluation component of Amodei's proposal. Accenture will help evaluate models, conduct red-teaming, and assess alignment and safeguards.
