OpenAI pushes mandatory national AI safety regulation

OpenAI frames its policy push as a response to a specific moment: what it calls a "defenders window," described by Greg Brockman, a limited period when frontier AI can help strengthen critical systems before powerful offensive capabilities become widespread. The company says capabilities are advancing quickly, pointing to its latest research showing AI agents can perform some tasks that would take skilled researchers several days. That is not recursive self-improvement, in its telling, but it is evidence of the direction of travel.

OpenAI's own safeguards are part of the context. It says it strengthened monitoring, alignment, and security across the model-development lifecycle, including universal monitoring of full trajectories and a mandatory alignment-evaluation gate before broader internal deployment for a system it calls Astra. It also says it will slow or stop development or deployment when proceeding would pose an unacceptable safety risk, as required per its preparedness framework. The policy ask follows that logic: if safeguards cannot keep pace with capabilities in individual labs, shared standards should.

The stakes, as OpenAI describes them, hinge on whether democratic governments can set durable rules before capabilities outpace the institutions meant to govern them. The company is betting that stronger safeguards will not slow adoption but enable it, and that compatible international standards will be necessary given how models and expertise move globally. Whether Congress acts before it adjourns—and whether states like California fill the vacuum in the meantime through what OpenAI calls reverse federalism—will show how much of that window remains open.