Fei-Fei Li calls for independent AI safety benchmarks

Fei-Fei Li, CEO of World Labs and Stanford University professor, is calling for an independent oversight system for advanced artificial intelligence (AI) evaluations. In recent interviews, Li argued that AI safety assessments should not be left solely to the companies developing the technology. She emphasized the need for a multi-stakeholder framework involving academia, government, and industry to establish common benchmarks and evaluation standards.

Li noted that if companies create their own criteria to evaluate their models, it becomes difficult to ensure objectivity and accountability. This discussion follows proposals from other industry leaders, such as Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, regarding granting external evaluators system access similar to company employees to inspect safety practices. However, critics like Cohere CEO Aidan Gomez have warned that allowing a small group of commercial interests to set AI rules could compromise independence.

Furthering the push for transparency, the AI Evaluator Forum has issued an open letter demanding that third-party agencies be granted independent judgment, editorial control, and sufficient system access without fear of corporate retaliation. While the debate continues regarding whether AI poses an existential threat or a manageable technical challenge, the central tension remains focused on who holds the authority to define evaluation standards and control the disclosure of results.