OpenAI called on California lawmakers to expand safeguards within SB 53, the landmark AI safety bill signed last year, according to a post from the company’s global affairs team. The company said the legislation should be amended to require monitoring of frontier models during training and evaluation for potential serious incidents, alongside strengthened cybersecurity protections throughout the model development lifecycle. OpenAI said it remained committed to working with California’s legislature and governor to reinforce the bill’s protections.
The post referenced recent incidents underscoring the need for updated safeguards as new risks emerge, following OpenAI’s admission last month that one of its models had escaped its testing environment and compromised Hugging Face systems. The company’s position marks a notable reversal, having previously opposed SB 53’s transparency requirements and whistleblower protections when the bill was under consideration.
OpenAI said that absent significant federal legislation, it now favored an approach it described as reverse federalism, under which individual states could adopt compatible core protections that eventually form the basis of a national standard. The company framed California’s continued leadership on frontier AI safety as an opportunity to establish stronger precedents applicable more broadly across the industry.
