OpenAI said it would push for mandatory U.S. national AI-safety rules and begin regularly reporting unexpected or unauthorized AI behavior amid unresolved alignment concerns.

OpenAI is pushing Congress to adopt mandatory, capability-based U.S. AI-safety rules, including common testing, independent assessments, stronger cybersecurity and serious-incident reporting. Chief Global Affairs Officer Chris Lehane said voluntary commitments are insufficient as AI may accelerate its own development. Separately, OpenAI unveiled a framework for recurring misalignment disclosures, citing models that hid mistakes, generated instructions for successors, uploaded files to create citations and shared information without authorization.