Anthropic said it has selected Accenture as an embedded evaluator to help test the safety of its artificial intelligence models, marking the first concrete step in CEO Dario Amodei’s proposal to slow the pace of advanced AI development.
According to the company, Anthropic and Accenture have agreed to invest at least $1 billion over the next five years to build capacity in this area. Anthropic said, however, that because of the “importance and urgency” of the issue, it will directly fund Accenture’s work for now. The company added that, over the longer term, it believes such efforts should be financed through pooled or government-backed sources, consistent with its Advanced AI Framework published in June.
Amodei recently outlined a three-step plan aimed at limiting how quickly AI companies improve their most advanced models. The first step calls for third-party evaluators to receive employee-level access so they can verify safety practices and report incidents. He said Anthropic had “unilaterally” committed to that part of the proposal and urged other AI developers to do the same.
Anthropic said the initial arrangement will place employees from Faculty, Accenture’s specialist AI business, inside the company to test safeguards, red-team models and assess whether those models behave in line with human values. The partnership is not exclusive, and Anthropic said it is also in discussions with research nonprofit METR and other third parties.
Anthropic and rival OpenAI have come under intense scrutiny in recent weeks after a growing number of researchers warned that AI could potentially cause catastrophic harm to humanity. Amodei’s proposal has drawn mixed responses across the tech sector. It was welcomed by OpenAI CEO Sam Altman and Tesla and SpaceX CEO Elon Musk, while Nvidia CEO Jensen Huang dismissed the concerns and argued that new regulation is unnecessary.
Questions remain over how Anthropic’s approach will work in practice, especially as the company prepares for what is widely expected to be a blockbuster IPO.
Anthropic stressed that it remains fully responsible for the safety of its models and said working with embedded evaluators will not reduce its accountability. The company said it is sharing these early steps so the public and other AI developers can see its process, adding that the approach is likely to evolve as the field matures.
