Anthropic isn’t having the best PR run of late.
Anthropic’s announcement that Accenture will independently evaluate its frontier AI models has drawn a Community Note on X. The note points out that Anthropic will pay for the work and already has a commercial partnership with Accenture.
Anthropic said the partnership will be led by Accenture’s specialist AI business, Faculty. It will cover red-teaming, alignment assessments and safeguard testing, and each company expects to invest at least $1 billion over five years. Anthropic says the arrangement is non-exclusive and that it will fund Accenture’s work directly.
The note argues that “independent” is a stretch. Anthropic funds the evaluator, and the two companies are already business partners.
Anthropic’s business ties to Accenture #
The relationship is substantial. In December 2025, the two announced a multi-year expansion that created the Accenture Anthropic Business Group and committed to training about 30,000 Accenture professionals on Claude. Accenture also became a premier partner for Claude Code, which it is rolling out to tens of thousands of its developers.
The deal included a joint offering for CIOs and co-developed solutions for regulated industries such as financial services, healthcare and the public sector. The companies have since launched Cyber.AI, a jointly built agentic cybersecurity platform. For Anthropic, Accenture is a major channel for selling Claude to large enterprises. That matters as its enterprise business keeps growing and the company reportedly prepares to go public.
Accenture isn’t exclusive to Anthropic. It announced a separate collaboration with OpenAI eight days before the Anthropic deal.
How Anthropic frames it #
The evaluation deal follows Dario Amodei’s essay calling for the industry to “pace the frontier”. Under “embedded evaluation,” outside evaluators work inside AI companies with access comparable to an employee’s. They can watch models in training, follow deployment decisions and talk directly to staff.
Anthropic addresses the funding question in its post. It says no settled funding system exists yet and that, long-term, funding should come from pooled or government sources. Until then, it plans to work with different evaluators under different funding arrangements. It says it is also in dialogue with METR and other nonprofit evaluators to pilot parts of the approach using their own funding. Questions, meanwhile, have also been raised about METR’s independence and its realtionshio with Anthropic. Anthropic adds that embedded evaluators don’t reduce its accountability, and that the safety of its models remains its responsibility.
Why the criticism has traction #
The timing raises the stakes. Anthropic recently disclosed that its models now lead 26% of its own AI R&D work. It also reported that Claude models gained unauthorized access to real organizations’ systems during cybersecurity tests. Credible oversight matters more when models are this capable.
One analysis notes the evaluators would have no authority to halt model development or deployment. Whether a paid evaluator with a large commercial relationship will publicly challenge its client remains to be seen. Anthropic itself says there are no standards yet for what evaluators can access or how they report findings.
The structure of the deal will be judged by what evaluators are allowed to see and publish, and by whether the additional nonprofit evaluators Anthropic has promised actually arrive.