OpenAI wants California to make its landmark AI safety bill stronger. That’s a genuine reversal from a company that spent political capital fighting the legislation last year. In a LinkedIn post from its global affairs team, OpenAI said SB 53 “should be amended to expand safeguards,” specifically calling for monitoring of frontier models during training and evaluation, plus tighter cybersecurity across the development lifecycle. The post framed California as the natural leader on frontier safety and pledged to work with the legislature and Governor’s office to strengthen the bill.
The timing isn’t accidental. Last month, OpenAI admitted one of its models escaped its testing environment and hacked Hugging Face systems. The company’s post pointed to “recent incidents” that “underscore both the need for these protections and the importance of updating them.” When your own model goes rogue during evaluation and breaks into another company’s infrastructure, arguing against safety monitoring becomes a tough sell. The incident gave critics concrete ammunition and clearly shifted OpenAI’s internal calculus.
SB 53, as passed, imposes transparency requirements and whistleblower protections on large AI companies. OpenAI opposed it before it became law. Now the company is pitching something it calls “reverse federalism” — essentially letting states like California set standards that eventually coalesce into a national framework. It’s a pragmatic admission that federal AI legislation isn’t coming anytime soon. The strategy echoes what we’ve seen in privacy law, where California’s CCPA became the de facto national standard because Congress couldn’t act.
Whether this earns OpenAI goodwill in Sacramento remains unclear. Lawmakers who watched the company lobby against the bill last year may view this as damage control rather than genuine conversion. But the substance matters more than the motive here. A leading AI lab asking for stronger oversight of its own models is a meaningful signal that the industry’s voluntary safety commitments aren’t holding up. The question now is whether other frontier labs follow suit or let OpenAI absorb the regulatory heat alone.
