The company that killed SB 1047 now wants tougher rules — but only the kind it can afford to follow.
The Summary
- OpenAI is urging California to strengthen SB 53, an AI safety bill the company previously opposed when it was called SB 1047
- The reversal comes after OpenAI spent 2024 lobbying against similar legislation, arguing it would stifle innovation
- The move signals OpenAI wants regulation written by incumbents, for incumbents — rules that lock in their lead while appearing responsible
The Signal
OpenAI's public stance flip on California AI regulation is not a road-to-Damascus moment. It is regulatory capture in real time. The company now supports strengthening SB 53, the rebranded successor to SB 1047, which it helped torpedo in 2024. The difference? OpenAI now has the resources, legal teams, and compliance infrastructure to meet whatever standards get written. Their smaller competitors do not.
This is the playbook Big Tech perfected with GDPR. Set the compliance bar high enough that only companies with armies of lawyers can clear it. Call it safety. Watch the market concentrate.
"When the biggest player asks for tougher rules, they are not asking for constraints. They are asking for moats."
SB 53 would require AI companies to conduct safety testing, disclose training data sources, and implement kill switches for models that pose catastrophic risks. Reasonable on paper. In practice, "catastrophic risk" remains undefined, "safety testing" has no agreed standard, and "kill switches" assume a level of control over emergent systems that nobody actually has. OpenAI knows this. They also know they can hire enough PhDs to write convincing safety reports while competitors scramble.
The timing matters. OpenAI is no longer the scrappy nonprofit that released GPT-2 with handwringing about misuse. It is a $157 billion company with Microsoft as a strategic partner and compute resources that dwarf most nation-states. Anthropic, Cohere, and Mistral are raising billions but still burning cash to train competitive models. Startups building agent frameworks on top of foundation models operate on venture timelines with single-digit millions in the bank.
Key points on who wins under strengthened AI safety bills:
- OpenAI, Google, Meta: compliance is a rounding error on their infrastructure spend
- Well-funded AI labs: can hire for it, slower but survivable
- Open-source projects and startups: dead in the water or forced into regulatory arbitrage by hosting elsewhere
California is not regulating AI. It is deciding which companies get to build it. OpenAI supporting stronger rules is not altruism. It is strategy. They want the drawbridge up before the next wave of competitors figures out how to scale.
The Implication
If you are building agent infrastructure, model tooling, or anything in the AI stack, watch what gets written into SB 53. The compliance costs will determine whether you can compete in California — which effectively means whether you can compete at all, given where the talent and capital sit. OpenAI's support should be a signal to price in regulatory moats as a feature, not a bug, of the next decade.
For everyone else: when the incumbent begs for regulation, the game is not safety. It is market design.