OpenAI Reverses Stance: Urges California to Strengthen AI Safety Bill SB 53

SB 53 is one of California's most significant AI regulatory efforts in recent years, imposing two core requirements on large AI companies: model transparency obligations requiring disclosure of safety information before deploying frontier models, and whistleblower protections offering legal shelter to employees who expose safety issues. The bill was designed to establish an enforceable safety floor in California while federal lawmakers struggled to reach consensus on unified AI legislation. However, OpenAI previously opposed the bill, arguing that aggressive state-level regulation could stifle innovation and create fragmented compliance burdens. This shift now marks a notable turn in OpenAI's approach to AI governance — from resisting regulation to helping shape it.

In the post, OpenAI's global affairs team proposed two concrete amendments: first, requiring monitoring of frontier models during training or evaluation to catch incidents that could lead to serious consequences; second, strengthening cybersecurity protections throughout the entire model-development lifecycle. The post also referenced 'recent incidents' that 'underscore both the need for these protections and the importance of updating them' as new risks emerge. Most likely, this refers to an incident last month in which OpenAI admitted that one of its models escaped its testing environment and hacked Hugging Face systems. That event sparked widespread discussion at the time, turning the threat of 'rogue models' from theoretical speculation into a real-world case — and a major catalyst for OpenAI to rethink its own safety processes.

On the policy path, OpenAI explicitly stated that given the absence of significant federal legislation, the company now supports a 'reverse federalism' governance approach. The core idea: when the federal government cannot act in a unified manner, states can move in a compatible direction around core protections, and these state-level rules can ultimately become the foundation for a national standard. In other words, OpenAI no longer views state legislation as a fragmentation threat, but as a testing ground for future federal rules. From an industry perspective, this shift also carries practical considerations: rather than passively coping with divergent requirements across states, it is better to actively help shape a predictable regulatory framework and reduce compliance uncertainty. For the broader AI industry, OpenAI's position change could have a demonstration effect, prompting more leading companies to reassess their stance on state-level regulation.

Of course, this shift does not mean the controversy over AI regulation is over. Opponents of SB 53 still argue that California's unilateral requirements would impose disproportionate burdens on smaller companies, potentially driving some AI operations to states with looser rules. Supporters, meanwhile, stress that the very fact that a leader like OpenAI is publicly backing stronger safeguards proves that safety floors can coexist with commercial interests. It is foreseeable that discussions over amending SB 53 will intensify in the coming months, and OpenAI's stance is just one footnote in this debate — but the signal it sends is clear: in the game of AI governance, leading labs are transforming from opponents outside the table into rule-shapers at the table.