Major AI Labs Co-Sign AEF-1 Standard for Third-Party AI Evaluations
Leading frontier AI companies—OpenAI, Anthropic, and xAI—have jointly co-signed the new AEF-1 standard for third-party AI evaluations. Published by the AI Evaluator Forum, AEF-1 provides a unified standard and checklist establishing operational baselines for external AI evaluators. As frontier AI models become increasingly powerful, self-evaluation by model developers is no longer sufficient to guarantee safety and regulatory compliance. Having major competing labs agree on a common evaluation framework helps establish trustworthy, independent auditing standards across the AI ecosystem. The AEF-1 standard outlines minimum operating conditions that evaluators must achieve regarding independence, model access, and evaluation transparency. Notably, the framework explicitly treats evaluations conducted under terms dictated by the AI developer as a lower standard of independence compared to fully independent audits.
## BACKGROUND
Evaluating artificial intelligence systems currently lacks standardized rules, making it challenging to consistently audit safety risks or verify capabilities across different organizations. Independent third-party evaluation relies on external entities assessing AI models to prevent bias and conflicts of interest. The AI Evaluator Forum introduced AEF-1 to formalize how these external auditors operate and report their findings.