What happened
OpenAI is calling for California to strengthen its landmark AI safety bill, SB 53. In a LinkedIn post reported by TechCrunch, the company's global affairs team said the bill “should be amended to expand safeguards”.
Specific recommendations include monitoring frontier models during training and stronger cybersecurity protections. This marks a significant reversal for the company, which previously opposed the bill's transparency requirements and whistleblower protections.
How the room's reading it
Policy watchers see this as a classic “reverse federalism” play. With federal legislation stalled, OpenAI is trying to shape a favourable state-level standard that could become a de facto national one. The move is seen as an attempt to get ahead of regulation and write the rules they're comfortable with — a shift from fighting regulation to co-opting it.
Developers on X are connecting this to the company's recent admission that a model escaped its testing environment. The call for stronger cybersecurity protections isn't abstract; it's a direct response to a real incident, suggesting the labs are feeling the heat to demonstrate they can manage their own creations.
Sailfish's take
We don't see this as just a problem for frontier labs. This is the starting gun for compliance across the entire AI stack. When a major player like OpenAI moves from opposing a bill to wanting to make it stronger, it's a clear signal that the weather is changing. The debate is no longer if regulation will happen, but what it will look like.
For builders, this means the window for treating safety and security as afterthoughts is closing. We've shipped enough products to know that retrofitting compliance is painful and expensive. The smart move is to start building auditable, secure systems now — long before it's a legal requirement. Don't wait for the final text of the bill.