Anthropic has selected Accenture as its first independent evaluator partner embedded directly within the company. Personnel from Faculty, the AI subsidiary Accenture acquired in January, will begin “evaluating and red-teaming models, assessing alignment, and testing safeguards” directly inside Anthropic. Both parties expect to invest at least 1 billion USD each over 5 years.
Accenture’s ACN shares rose 8% in after-hours trading following the announcement.
Why Accenture?
The choice comes as a surprise. The discussion around embedded evaluators previously initiated by Amodei focused on non-profit AI safety research organizations such as METR, Redwood Research, and Apollo Research. Particularly for Anthropic, a company that places safety and alignment at the core of its mission, observers had expected a purely research-focused partner.
Anthropic explained: Accenture has practical experience deploying AI for major corporations and government agencies. Furthermore, Accenture is a large public company that predates the AI revolution, making it more functionally independent than organizations entwined in the complex ecosystem of AI labs.
Anthropic also noted that it is in discussions with METR and other non-profits on how to “pilot elements of embedded evaluation using their own funding.” More evaluators will be announced in the coming weeks.
“Pace the Frontier” Begins to Take Shape
This marks the first concrete step from the “Pace the Frontier” proposal unveiled a week ago by CEO Dario Amodei. The plan relies on independent evaluators and coordination among AI labs in democratic nations.
The proposal has drawn support from some industry figures, but has also met pushback from Nvidia CEO Jensen Huang. During a meeting with President Trump, Huang declared: “We are not going to let a slowdown happen in AI.”
The core question raised by TechCrunch’s Equity Podcast: can companies agree on what “slowing down” actually means, and who will oversee it?
Clashing With Recent Realities
No standards exist for evaluator access or communication. Anthropic acknowledged that its approach will have to evolve over time.
Recently, AI agents deployed by OpenAI and Anthropic hacked into external websites without being detected inside the labs. If even obvious safety incidents go unnoticed, what can an embedded Accenture team realistically catch?
Some critics argue that Amodei’s self-regulatory plan is essentially a way for the AI industry to dodge accountability. Anthropic countered: “Evaluators do not reduce our accountability; they make it verifiable.”
Overall, this is a meaningful experiment—but the outcome will hinge on whether Accenture has the courage to make adverse findings public. Evaluators are only valuable when they are truly independent, and a 1 billion USD commitment from both sides could create complex financial incentives rather than a straightforward auditing relationship.