Fei-Fei Li, CEO of World Labs, stated on September 22, 2026, that AI evaluation should not be controlled solely by the companies developing the technology. She advocates for a multi-stakeholder approach involving academia, government, and industry to establish independent benchmarks for frontier AI systems.
Her comments follow a letter from the AI Evaluator Forum, a group of over 100 experts, which criticized recent commitments from Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman. The coalition argued that providing third-party evaluators with employee-like access to models is insufficient to ensure structural independence.
Li previously co-led the California Joint Policy Working Group on AI Frontier Models in March 2025. That group proposed policy recommendations for AI labs, including mandatory public reporting of safety test results, enhanced third-party evaluation standards, improved data practices, and whistleblower protections.
The proposed framework aims to create a permanent evaluation infrastructure. Li suggests that academia should provide methodological rigor, government should ensure democratic accountability, and industry should contribute technical expertise, preventing any single entity from dominating the assessment process.
