OpenAI urges U.S. lawmakers to move quickly on frontier AI safety rules

OpenAI has called for faster government action on frontier AI safety, arguing that the current moment offers a narrow policy window before advanced model capabilities outpace public institutions. In a September 9, 2026 post by Chief Global Affairs Officer Chris Lehane, the company said that AI policy needs to move with the pace of model progress and that technical safeguards inside individual labs will not be enough.

The post is notable because it comes from one of the companies most directly involved in developing frontier systems. OpenAI says it wants to work with Congress on mandatory, capability-based national AI safety regulation. The company describes a federal framework that could include common testing, independent assessments, stronger cybersecurity protections, incident reporting, national preparedness and shared measures for tracking progress toward AI systems that can accelerate AI development itself.

OpenAI is also backing state action while Congress considers broader rules. The company said it now supports four California bills that have passed the legislature and are headed to Governor Gavin Newsom. The measures cover independent AI risk assessors, AI auditor standards, protections for young people using companion chatbots, and screening rules for gene-synthesis providers and makers of benchtop synthesis equipment. OpenAI said some of these proposals were not ones it had supported in the past, but that recent capability gains changed its assessment.

The company's framing reflects a changing policy debate. For years, AI regulation was often split between advocates who warned of long-term catastrophic risk and companies that argued broad rules could slow innovation. OpenAI is now presenting safety rules as a prerequisite for trust and adoption. Its post says frontier requirements should apply to the handful of well-resourced labs developing the most capable systems, while avoiding burdens on startups, smaller developers or researchers far from the frontier.

Monitoring is another major theme. OpenAI says industry standards should require developers to watch for misaligned behavior as models gain autonomy, use more powerful tools and operate over longer periods. It also supports disclosure requirements when a model, during development or evaluation, bypasses another organization's security controls and materially accesses, changes or destroys protected systems or confidential information. The company says its own work on reporting consequential misalignment incidents could inform broader policy.

The post stops short of claiming that autonomous recursive self-improvement is happening now. OpenAI says it is not, but argues that AI is already accelerating parts of research used to build and align future models. That distinction matters: the company is warning that the path toward more capable systems is visible enough for governments to set measuring methods, human-control requirements and safety bars before the hardest risks arrive.

The larger significance is that OpenAI is trying to shift the conversation from voluntary commitments to enforceable rules. Its position will still face skepticism from critics who worry about regulatory capture or vague standards. But the message is clear: one of the leading AI labs is asking lawmakers, states, other labs and international partners to move faster, because the governance debate is no longer separate from the product race.