OpenAI calls for mandatory AI safety laws and backs California bills

Big tech has spent decades treating government regulation like a plague, offering vague voluntary pledges just to keep lawmakers at arm’s length.

That playbook is officially out the window. OpenAI just published a call to action by Chief Global Affairs Officer Chris Lehane, urging the U.S. Congress to pass mandatory, capability-based national safety rules for frontier AI before the legislative window shuts.

While Congress drags its feet, OpenAI is actively backing four California state safety bills currently sitting on Governor Gavin Newsom’s desk. OpenAI calls this strategy reverse federalism: letting individual states set the baseline until the federal government is forced to step in and codify it nationwide.

Four state bills and the Astra signal

The four California bills OpenAI endorsed target concrete, immediate vulnerabilities.

SB 813 sets up infrastructure to certify qualified, independent bodies to assess high-end AI risks. AB 1405 establishes strict registration, transparency, and accountability rules for the AI auditors running those checks.

On the consumer front, SB 1119 targets companion chatbots, demanding age assurance, parental controls, and independent audits to protect kids and teens. Finally, AB 1864 targets gene-synthesis providers and equipment makers to prevent frontier models from enabling biological threats.

Why the sudden rush? OpenAI pointed straight to its own research floor. The post cited capabilities seen in its upcoming Astra system and a warning from Chief Scientist Jakub Pachocki regarding recursive self-improvement.

While fully autonomous self-improving AI does not exist today, OpenAI says AI agents are already accelerating its own internal research, completing tasks in days that previously took skilled researchers several days to solve.

To manage Astra, OpenAI had to roll out universal monitoring across all execution trajectories, including raw chains of thought, alongside a mandatory internal safety gate before deployment. In short: internal progress is moving so fast that internal guardrails alone are no longer enough.

Why it matters

We think this is a masterclass in regulatory strategy, cloaked in public-spirited concern.

On the surface, demanding mandatory rules looks like OpenAI voluntarily placing handcuffs on itself. But look at where they want the line drawn: Lehane explicitly argued that heavy requirements should only hit the tiny handful of well-funded labs operating at the bleeding frontier, exempting open-weight models, researchers, and smaller startups.

By pushing for mandatory independent audits and federal oversight gates, OpenAI effectively helps design compliance standards that only well-capitalized tech giants can afford. It protects open-source from direct bans, but ensures that anyone trying to cross the boundary into frontier capabilities faces a massive regulatory moat.

For enterprise users and developers, this means commercial AI deployment will quickly become more standardized and legally defensible. But the risk is clear: when the most dominant players get to dictate the terms of their own oversight, the line between public safety and incumbent protection gets blurry fast.