OpenAI Pushes California to Expand AI Safety Law It Backed
The company wants SB 53 amended to require monitoring for security bypass attempts during model training and evaluation.
OpenAI is calling on California lawmakers to strengthen the state's AI safety legislation less than a year after it became law, proposing new monitoring requirements aimed at detecting when frontier models attempt to bypass security controls.
In a LinkedIn post from its Global Affairs team, OpenAI outlined amendments it wants made to SB 53, the Transparency in Frontier Artificial Intelligence Act that Governor Gavin Newsom signed in September 2024. The company is specifically requesting that the law mandate monitoring of frontier AI models during training or evaluation for signs they could circumvent third-party security controls or access confidential information without authorization.
OpenAI is also advocating for enhanced cybersecurity protections throughout the entire model development lifecycle, focused on preventing AI systems from bypassing their own internal safeguards.
Why it matters
The push reveals how quickly AI capabilities are evolving and how even recently enacted regulations may struggle to keep pace. OpenAI's willingness to advocate for stricter oversight of its own industry—while positioning itself as a key voice in shaping those standards—reflects both genuine safety concerns and strategic positioning as states fill the federal policy vacuum.
The 'reverse federalism' strategy
OpenAI frames its approach as "reverse federalism," arguing that while Congress remains deadlocked on comprehensive AI legislation, states can establish compatible safety standards that eventually form the foundation for national policy. The company emphasized that harmonization shouldn't mean freezing requirements in place as the technology rapidly advances.
"The goal is not to write rules for one particular incident but to build a framework that helps developers detect problems earlier, respond quickly, and share lessons that can improve safety across the industry," OpenAI stated in its post.
What SB 53 already requires
The existing law already represents significant regulatory territory. According to the Governor's office, SB 53 requires large frontier AI developers to publish safety frameworks, establishes formal incident reporting channels to California's Office of Emergency Services, protects whistleblowers, and gives the state Attorney General authority to impose civil penalties for non-compliance.
The legislation also created CalCompute, a public computing consortium designed to support safety and equity research, and mandates annual reviews of the law to account for technological changes. Newsom positioned the bill as evidence that California could protect public safety without stifling the AI industry that drives much of the state's economy.
Timing and context
While OpenAI characterized its proposal as forward-looking rather than reactive to any specific incident, the timing comes as the AI industry faces mounting scrutiny over safety practices and the adequacy of existing safeguards. The company maintained its amendments would help developers detect problems earlier and ensure lessons are shared industry-wide.
These details were first reported by Mashable.
This is an original analysis by the Omega editorial team. Source reporting: AI Watch.
Want systems like this working for your business?
Book a Call
