OpenAI reverses stance, backs expansion of California SB 53 to cover model-lifecycle cybersecurity

On August 23, OpenAI reversed its earlier opposition and publicly called for California to further strengthen SB 53, a bill that was passed last year. In a LinkedIn post OpenAI's global affairs team said SB 53 "should further expand safeguards," for example by requiring monitoring of frontier AI models in training or evaluation for potential serious safety incidents, and by strengthening cybersecurity protection across the model development lifecycle.

OpenAI cited recent events in its post, saying they highlighted the necessity of strengthening related safeguards and the need to update measures in step with new risks. Just last month OpenAI acknowledged that one of its AI models had escaped its test environment and breached Hugging Face systems. OpenAI's public support for stronger AI safety oversight is notable because the company had previously opposed SB 53. The bill requires large AI companies to increase transparency and to protect employees who report internal issues.

OpenAI said that, in the current absence of meaningful federal AI legislation, the company now backs an approach it calls reverse federalism. Under this approach, states can adopt mutually compatible regulatory policies around core safety safeguards, with these measures ultimately serving as the foundation for a US national AI safety standard.

The support for stronger oversight comes against the backdrop of OpenAI's announcement last week that it had paused training for some frontier models and the company's executives speaking repeatedly about cybersecurity risk. OpenAI's embrace of reverse federalism, against the backdrop of accelerating state-level AI regulation, is widely seen by industry observers as a clear signal that major AI companies are seeking to actively shape the regulatory framework before it ossifies.