Anthropic taps Accenture as its first embedded safety evaluator

Anthropic said staff from Accenture, the technology consulting giant, will begin working inside the company to scrutinize its models and staff — a first step in Dario Amodei’s plan to put third-party safety evaluators inside AI labs. In a blog post, Anthropic said Faculty, a company Accenture acquired in January to act as its AI division, will start evaluating and red-teaming Anthropic’s models, running alignment assessments, and testing model safeguards. Both companies expect to invest at least $1 billion in the project over the next five years.

The choice of Accenture surprised many AI watchers — and the markets, where the consulting firm’s shares jumped 8% after hours. The discussion that followed Amodei’s original blog post on embedded evaluators had focused on AI safety research organizations like METR, Redwood Research, and Apollo Research — particularly at Anthropic, which places AI safety and alignment at the heart of its mission. Anthropic said more evaluators will be announced in the weeks ahead, and that it is in talks with METR and other nonprofit organizations about piloting elements of embedded evaluation funded by the nonprofits themselves.

While Accenture isn’t known for work on the cutting edge of deep-learning research, Anthropic pointed to the company’s practical experience deploying AI for large corporations and government agencies as a key advantage. As a large public company that predates the AI boom, Accenture is also more functionally independent of Anthropic and the ecosystem surrounding the lab, Anthropic said. The lab noted that no standards yet exist for evaluators’ access or communications, and that it expects its approach to evolve over time.

External evaluations are already a major part of the release process for new large language models, but recent incidents have raised the stakes: AI agents deployed by OpenAI and Anthropic have hacked into outside websites without raising alarms inside the labs. Some critics of the industry see Amodei’s scheme for self-policing as a plan to evade accountability for AI models’ misbehavior. Anthropic disagrees: the evaluators, it said, “do not reduce our accountability, but help to make it more verifiable. The safety of our models remains our responsibility.”

Why it matters

Whether embedded evaluators like Faculty can credibly police the AI labs that fund them — rather than simply lend cover to self-regulation — will shape how regulators and rivals judge Anthropic’s push to make AI safety oversight look independent.