In a Friday statement, Anthropic said Accenture will embed a dedicated team inside the AI lab to red‑team its latest frontier models, conduct alignment assessments and test safeguards, giving the evaluators access comparable to Anthropic’s own staff.
The partnership marks the first concrete implementation of a commitment outlined in Dario Amodei’s essay released six days earlier, which called for independent, embedded evaluation of advanced AI systems.
Anthropic will fund Accenture’s work directly for now, but the company emphasized that long‑term financing should come from pooled or government sources – a model it says does not yet exist.
Accenture is already Anthropic’s largest Claude Code deployment, with roughly 30,000 of its professionals trained on the model and tens of thousands of developers using Claude Code, making it a natural choice for the role, according to Anthropic.
The two firms each expect to invest at least $1 billion over the five‑year term, a sum that Anthropic argues is needed to staff a standing evaluation team, something smaller nonprofit evaluators such as METR cannot currently sustain.
Anthropic acknowledged that there are currently no industry standards governing what embedded evaluators may access or how they must report findings, noting that the arrangement places the evaluator under the same governance framework as the company being examined.
The collaboration is non‑exclusive; Anthropic said additional evaluators will be added within weeks, and Accenture will perform similar work for other AI developers.
While the partnership expands Anthropic’s internal safety capabilities, observers note the broader challenge that independent evaluation often ends up being owned by parties with commercial stakes, raising questions about the objectivity of any critical findings that could delay product releases.