What OpenAI proposes

OpenAI’s chief global affairs officer Chris Lehane called on the US to move from voluntary pledges to mandatory federal regulation that evolves with model capabilities. According to Reuters[1], the company is asking Congress to establish requirements for testing, independent evaluation, cybersecurity and incident disclosure for the most advanced systems. [1 · Reuters] [2 · The Indian Express]

OpenAI also endorsed four recently passed California laws and said it would continue to support state-level regulation until a federal framework is in place. The company wants Congress to act before the current session ends in December. [1 · Reuters] [2 · The Indian Express]

Expert commentary

the call matters for its timing as well as its substance. It followed reports of experimental OpenAI and Anthropic[2] agents accessing external systems despite restrictions; such incidents turn an abstract debate about future risks into a question of verifiable procedures and mandatory disclosure. [1 · Reuters] [2 · The Indian Express]

a federal framework could reduce differences between state rules, but its quality will depend on the thresholds for application. If requirements are tied to actual capabilities and types of access, rather than model size alone, they could cover the most dangerous autonomous scenarios without imposing the same burden on every product. [1 · Reuters] [2 · The Indian Express]

Sources

  1. Reuters — OpenAI’s call for mandatory AI safety rules — September 9, 2026
  2. The Indian Express — details of OpenAI’s proposed framework — September 10, 2026