OpenAI is making a decisive pivot from voluntary guidelines to hard regulatory advocacy. Chris Lehane, the company’s Chief Global Affairs Officer, argues that the current pace of AI development demands immediate policy action rather than waiting for legislative perfection. The tech giant is actively supporting four specific bills in California: SB 813 for independent safety assessments, AB 1405 for auditor standards, SB 1119 for youth protections, and AB 1864 to guard against biological threats. This state-level maneuvering serves as a stopgap while OpenAI pushes Congress for mandatory, capability-based national safety regulations. It is a clear signal that the industry leader believes self-regulation is no longer sufficient to manage the risks of frontier models.
The urgency stems from what Greg Brockman calls a “defenders window.” This is a limited period where defensive systems can be strengthened before offensive AI capabilities become widespread and difficult to contain. Lehane emphasizes that advanced AI offers massive benefits, from accelerating medical research to fortifying critical infrastructure. However, these same capabilities pose existential risks if they fall into the wrong hands or spiral out of control. With open-source models rapidly approaching frontier performance, the technology will not remain confined to a few secure laboratories. The window to establish durable safeguards is closing, and policymakers must act before institutions are overwhelmed by the speed of innovation.
Internally, OpenAI is tightening its own controls. The company has introduced universal monitoring of full reasoning trajectories and mandatory alignment-evaluation gates for its Astra model. Chief Scientist Jakub Pachocki warns that recursive self-improvement requires “extreme caution,” noting that while fully autonomous self-improving AI isn’t here yet, agents are already accelerating research tasks that once took days. OpenAI commits to slowing or stopping development if safety risks become unacceptable. This stance challenges the common narrative that safety measures stifle innovation. Instead, Lehane posits that confidence in safety is the actual engine of sustainable progress.
This shift places OpenAI in a complex position. By advocating for strict international standards and potential development pauses, they are effectively lobbying for rules that could cement their market dominance while raising barriers for competitors. The call for shared global standards on when to slow down AI development is particularly striking. It suggests a future where progress is not just measured by speed or capability, but by verified safety metrics. Whether governments can coordinate quickly enough to implement these frameworks remains the biggest variable. The technology is moving faster than the bureaucracy designed to govern it.
