More than 100 AI experts signed a public letter on September 18, 2026, issued by the AI Evaluator Forum (AEF). The document demands that third-party evaluators of frontier AI systems maintain genuine independence from the companies they assess, rejecting the model of embedding evaluators within corporate structures.
The AEF letter directly challenges proposals from Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, who suggested a 'nested' model where evaluators receive employee-like access. The forum argues this creates a structural conflict of interest that compromises the integrity of safety testing.
The forum outlines five non-negotiable conditions for credible evaluation: structural independence, diverse viewpoints, public transparency of findings, protection against retaliation for evaluators, and full access to systems and data. AEF chair Conrad Stosz emphasized that these are baseline requirements.
The initiative builds on the AEF-1 standard, launched by the forum in December 2025. While companies like Anthropic and Google have begun implementing parts of this standard, the AEF notes that current corporate proposals for internal oversight remain at odds with their framework.
Adopting rigorous, independent evaluation practices may influence institutional capital allocation. Companies that prioritize verifiable governance are viewed more favorably by investors, while those resisting such standards face potential reputational risks and regulatory penalties.
