Anthropic Selects Accenture as 'Independent Evaluator' for AI Safety, Funding Its Own Audits
Anthropic has introduced Accenture as an independent evaluator for its AI models, despite the firm being funded by Anthropic and holding long-term contracts with the company.
Anthropic Selects Accenture as 'Independent Evaluator' for AI Safety, Funding Its Own Audits
Anthropic has announced the first "independent evaluator" under its Embedded Evaluators program, a concept previously promoted by Dario Amodei in his essay on slowing down AI development.
The chosen company is Accenture. Its mandate includes conducting:
- Red-teaming exercises;
- Alignment checks;
- Testing of safeguards for Claude models.
However, the "independent" status of Accenture raises significant questions:
- Accenture trains approximately 30,000 of its employees on Claude.
- The two companies have a multi-year contract dating back to the end of 2025.
- Most critically, Anthropic explicitly states that it will be funding Accenture's work in this capacity.
This mirrors a previous situation with METR, which was granted "independent access" by Anthropic to investigate agent incidents, despite METR also being indirectly financed through Anthropic's equity.
In essence: Anthropic claims to bring in independent evaluators but selects them from a pool of business partners it directly finances.
Why it matters
AnalysisThis development undermines trust in AI safety audits, as 'independent' reviewers are financially dependent on the companies whose models they are evaluating, creating a clear conflict of interest.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
AI Alignment Problem
A fundamental safety issue in artificial intelligence: aligning the goals and motivations of autonomous systems with human values, ethics, and the preservation of life. It explores the risks of a superintelligent system finding a literal or destructive way to fulfill poorly defined tasks.
Claude rejected a request because of the word “cyber”
A brief observation that Claude refused a task due to the word “cyber”.
Claude Haiku (Fast Compact Models by Anthropic)
The most compact and fastest model in the Anthropic lineup, designed for lightning-fast responses, mass text classification, data extraction, and agent routing at minimal cost.
Service estimates the risk of being banned by Anthropic
Chinese developers launched a tool that analyzes IP address, country, ISP, connection type, DNS and WebRTC leaks to score trust from 0 to 100 for Anthropic-related access.