AI experts urge Anthropic and OpenAI to adopt independent safety evaluators

SAN FRANCISCO — A coalition of over 100 artificial intelligence experts published an open letter on Sept. 18, 2026, calling for foundation model laboratories including Anthropic and OpenAI to establish truly independent safety evaluators for their systems, according to a report by CNBC Technology.

Strategic Context

The public letter targets the current evaluation structures utilized by leading artificial intelligence developers. Up to this point, much of the safety testing and security assessment for frontier models has been conducted internally or through arrangements closely managed by the developers themselves. The coalition of more than 100 signatories argues that this setup creates inherent conflicts of interest and lacks the structural independence necessary to verify the safety and security risks of increasingly capable models.

The demand for independent oversight arrives as foundation model developers push deeper into enterprise and government deployment, where third-party validation of model capabilities and failure modes carries significant regulatory and operational weight. By pressing for greater transparency and distinct separation between developers and evaluators, the signatories aim to establish a standardized baseline for how model safety is measured before deployment.

Forward Outlook

Operators, enterprise buyers, and allocators monitoring the artificial intelligence sector will need to watch how labs respond to these calls for structural independence. As procurement cycles increasingly factor in safety compliance, third-party verification models could shift from voluntary frameworks to mandatory benchmarks for commercial deployment.