Anthropic and Accenture Ink Landmark $2B AI Safety Agreement: An Industry First
Anthropic, the AI safety firm responsible for the Claude family of models, declared on Friday, September 18, its selection of the global consulting behemoth Accenture as its inaugural embedded evaluator. This decision marks a significant stride in the quest to enhance transparency and accountability in frontier AI development.
As per the agreement, Accenture’s specialist AI division, Faculty — which Accenture acquired in 2025 — will station staff inside Anthropic with employee-level access. Their mission includes red-teaming AI models, conducting alignment assessments, and stringently testing the effectiveness of existing safeguards. Both firms have pledged a minimum investment of $1 billion over five years, culminating in a combined commitment of $2 billion, with Anthropic directly financing Accenture’s work.
This collaboration represents the initial tangible move in a wider three-phase strategy unveiled by Anthropic CEO Dario Amodei on September 12. The strategy aims to intentionally decelerate the rate of frontier AI development while concurrently fortifying safety infrastructure. Amodei has expressed that Anthropic has “unilaterally” committed to the embedded evaluator model and has urged other AI labs to do the same.
The agreement is non-exclusive, with Anthropic confirming ongoing negotiations with AI safety research nonprofit METR and other third parties for similar collaborations. The firm stressed that partnering with external evaluators does not diminish its own responsibility for model safety.
Source: CNBC – Anthropic selects Accenture as first embedded evaluator in safety push