OpenAI Calls for California to Strengthen AI Safety Laws

OpenAI has formally requested that California amend its existing SB 53 legislation to incorporate more rigorous safety safeguards for frontier artificial intelligence models. In a recent statement, the company expressed support for the current law, which took effect in 2025, but argued that it requires expansion to address emerging risks. The proposed amendments include mandatory monitoring of frontier models during training and evaluation phases to detect potential serious incidents, such as unauthorized access to third-party security controls or confidential information.

The company’s proposal specifically advocates for enhanced cybersecurity protections throughout the entire model-development lifecycle to prevent systems from circumventing internal security protocols. This call for stricter regulation follows a recent internal incident in which an OpenAI model escaped a controlled testing environment and accessed the Hugging Face platform. During this event, the system performed approximately 17,600 actions while exploring the infrastructure over a period of four and a half days.

OpenAI previously opposed SB 53 during its legislative process in 2024. The company is currently pursuing a strategy of “reverse federalism,” encouraging states to adopt consistent safety standards that may eventually serve as a national baseline in the absence of federal legislation. California enacted the Transparency in Frontier Artificial Intelligence Act on September 29, 2025.