Anthropic picks Accenture for third-party AI safety evaluations

Anthropic Partners with Accenture for Third-Party AI Safety Evaluations

This move represents the first step in the CEO’s plan to slow down AI development.

Anthropic announced a partnership with Accenture to serve as third-party evaluators for the company’s frontier AI models. The collaboration will embed Accenture’s evaluators within Anthropic to assess safety compliance and model alignment. Both companies will each invest at least $1 billion over the next five years in this joint effort.

CEO’s Three-Step Plan to Slow AI Development

The partnership aligns with CEO Dario Amodei’s recent proposal to decelerate AI advancement. His plan calls for all “frontier AI” companies to provide “ongoing, employee-like access” to embedded third-party evaluators. These evaluators are tasked with:

1. Ensuring safety practices and commitments are followed.
2. Reporting incidents.
3. Assessing AI model alignment.

Scope of Accenture’s Evaluator Role

Accenture’s evaluators will monitor models during training, oversee the decisions governing model construction and deployment, and communicate directly with employees. Anthropic highlighted Accenture’s experience in helping businesses and governments deploy AI across industries, noting that the consulting firm can offer insights into enterprise AI usage and its implications for Anthropic’s models.

Challenges and Future Steps

Anthropic acknowledged that embedded evaluators are a novel concept, with no established standards for scope, evaluator access, or reporting procedures. To address this, the company plans to collaborate with additional evaluators beyond Accenture. Anthropic stated it will announce other evaluators in the coming weeks and emphasized that Accenture will also work with other AI companies.