On September 9, 2026, Chris Lehane, OpenAI’s Chief Global Affairs Officer, argued that a new phase in AI capabilities requires faster policy action. OpenAI is pushing for mandatory, capability-based national regulation and said no single actor can manage frontier risks alone.
Federal push and state steps
OpenAI reiterated support for a federal framework that would set common testing, independent-assessment requirements, cybersecurity protections, incident-reporting rules, and measures to track progress toward possible recursive self-improvement. The company cited its “Blueprint for Democratic Governance of Frontier AI” as a path toward such a framework and said it will engage with Congress on legislation that materially raises safety requirements.
While urging Congress to act before it adjourns, OpenAI also said it will back state-level measures. The company formally endorsed four California bills headed to Governor Newsom: SB 813 (process for designating qualified, independent organizations for AI risk assessment), AB 1405 (registration and standards for AI auditors), SB 1119 (age assurance, risk assessments, independent audits, parental controls and safeguards for youth-facing chatbots), and AB 1864 (requiring gene-synthesis providers and benchtop equipment makers to follow federal screening standards).
Standards, monitoring and technical safeguards
OpenAI called for industry-led standards to monitor frontier models and detect misalignment—defined as models pursuing objectives that violate human intent without needing conscious intent. The company is developing a framework for reporting consequential misalignment incidents and systematic monitoring of frontier-model activity, and it said those company-led efforts should inform federal reporting rules.
OpenAI attributed warnings about rapid capability growth to its Chief Scientist Jakub Pachocki, who urged “extreme caution,” and quoted Greg Brockman’s concept of a limited “defenders window” when defenders can harden systems before offensive uses spread. The company said fully autonomous recursive self-improvement is not happening today and should not be pursued unless it can be done safely.
OpenAI described internal safeguards already in place—stronger isolation for frontier workloads, expanded monitoring during tool-enabled training, a mandatory alignment-evaluation gate for Astra before wider deployment, and a preparedness framework that requires slowing or stopping development when risks are unacceptable. The company said it will continue to support legislation and industry standards that raise the safety bar.
Original source: OpenAI News