OpenAI Flips on California AI Safety Bill, Now Pushing to Make It Tougher

OpenAI has dramatically shifted its stance on California’s high‑stakes AI safety bill SB 53. Tech media report that the company, which previously lobbied against the proposal, is now urging lawmakers to strengthen the legislation rather than water it down.

SB 53 targets so‑called “frontier models” – the largest and most capable AI systems – and would require developers to conduct formal safety evaluations and submit concrete incident‑response and containment plans to state authorities. The reversal comes just as a new study by Guidelight AI Standards graded leading labs such as Anthropic, Google, OpenAI, Meta and xAI on how well they could contain a misaligned or “rogue” model. While OpenAI scored comparatively well, the report concluded that none of the firms have sufficiently transparent, publicly documented playbooks for worst‑case scenarios.

If California expands and passes SB 53, it could become a de facto benchmark for AI safety governance in the United States, much as state privacy laws influenced national data rules. For AI companies, the debate highlights a growing tension: how to move fast enough to stay competitive in frontier research while accepting more intrusive oversight and disclosure about risks, testing and emergency shutdown procedures.