Specify mandatory requirements for frontier AI developers

Determine which specific safety, security, reporting, oversight, and risk‑governance requirements should be imposed on developers of frontier artificial intelligence systems and delineate their scope and conditions of applicability.

Background

The chapter surveys candidate safety practices (e.g., evaluations, security controls, reporting, external scrutiny) that could be mandated for developers of highly capable general‑purpose AI systems. It emphasizes that the risk landscape is evolving and many practices are nascent, making it difficult to fix a definitive set of requirements at present.

Clarifying which requirements should be imposed—and under what conditions—would guide both policymakers in drafting legislation and developers in implementation, and would reduce uncertainty while aligning industry behavior with public safety objectives.