Anthropic has selected Accenture as its first embedded evaluator to monitor safety and implement CEO Dario Amodei's proposal to slow the pace of frontier AI model development.
Key facts
- •Anthropic and Accenture are each committing at least $1 billion over five years to support the evaluation initiative.
- •Accenture’s specialist AI business, Faculty, will lead the embedded evaluation process.
- •The partnership allows evaluators to watch models in training, speak directly to employees, and report incidents.
- •Anthropic CEO Dario Amodei proposed the three-step plan for slowing AI development on September 19, 2026.
- •Anthropic remains responsible for the safety of its models despite the use of third-party evaluators.
Anthropic announced on Friday that it has partnered with Accenture to conduct independent safety evaluations of its frontier AI models. This initiative marks the first step in implementing a three-step plan proposed by Anthropic CEO Dario Amodei to temper the speed of AI development. Both companies have committed to investing at least $1 billion each over the next five years to build capacity for this work.
Embedded Evaluation Framework
Under the partnership, Accenture’s specialist AI business, Faculty, will embed evaluators directly within Anthropic. These evaluators will have employee-level access to the company, allowing them to monitor how models are built and deployed, conduct red-teaming, and assess whether models align with human values. Anthropic stated that this vantage point will help verify safety commitments and identify potential blind spots.
Industry Context and Future Plans
The partnership follows a proposal by CEO Dario Amodei on September 19, 2026, which called for AI companies to grant independent evaluators greater access to their systems. While Anthropic is funding Accenture’s work directly for now, the company stated that it believes long-term funding should eventually come from government or pooled sources. Anthropic noted that the partnership is not exclusive and that it is in discussions with other organizations, including the research nonprofit METR, to serve as additional evaluators.
Timeline
- September 17, 2026OpenAI announced it would begin publishing reports on concerning model behavior.
- September 19, 2026Anthropic CEO Dario Amodei published a three-step plan to slow frontier AI development.
- Friday, September 25, 2026Anthropic announced its partnership with Accenture for embedded evaluation.
Advertisement
This article was independently rewritten by ManyPress editorial AI from reporting originally published by The Hindu Technology, CNBC Technology, Engadget.
