Anthropic and Accenture Invest $2 Billion in AI Safety
Anthropic and Accenture are investing $2 billion to establish embedded third-party safety evaluations for frontier AI models amid intensifying competition with OpenAI.
AI laboratory Anthropic and professional services firm Accenture have entered a five-year partnership to invest at least $2 billion into the independent evaluation of frontier AI models. Through Accenture's AI business, Faculty, the initiative implements embedded evaluation, granting independent assessors employee-level access to internal systems, data, and physical spaces to conduct red-teaming and alignment assessments. This non-exclusive arrangement allows evaluators to publish findings without Anthropic's editorial control, though the company may redact security-sensitive or legally privileged information.
The partnership follows a September 18 demand from the AI Evaluator Forum, a coalition of over 100 experts including Geoffrey Hinton, for standardized principles to ensure scientific objectivity and independence for third-party evaluators. While OpenAI and Microsoft CEOs have expressed support for the embedded model, the Forum warned that oversight lacks credibility without robust protections against corporate interference.
These safety efforts coincide with extreme market pressure. Anthropic is weighing the release of a new Claude model to counter OpenAI's GPT-6 Astra, which has captured a larger share of enterprise AI spending. Despite this, Anthropic reported a July annualized revenue run rate exceeding $65 billion, surpassing OpenAI's $40 billion. The company is also balancing its development speed—with Claude now assisting in 26% of its own R&D—against calls from CEO Dario Amodei to slow frontier model releases due to systemic risks. Consequently, Anthropic is considering postponing its initial public offering until after the November US midterm elections.
