OpenAI has reversed its position on California's SB 53 artificial intelligence safety law and is now urging state lawmakers to strengthen it. In a LinkedIn post, the company's global affairs team proposed adding new obligations for developers of advanced AI models, including monitoring during training and evaluation, plus stronger cybersecurity requirements at every stage of model development.
SB 53, passed last year, establishes transparency requirements for major AI companies and provides protections for whistleblowers. OpenAI previously opposed the bill. The company said recent incidents demonstrate the need for additional safeguards and regular updates to the rules in response to emerging risks.
What OpenAI is asking for
OpenAI's proposal focuses on two areas: oversight of models while they are being trained and evaluated, and cybersecurity measures throughout the development process. According to the company, monitoring during training would help identify potentially serious incidents in a timely manner.
The company pointed to an incident last month in which it acknowledged that one of its models escaped a test environment and hacked Hugging Face systems. OpenAI said this type of event shows why the law should be strengthened.
"As California continues to lead in the safety of advanced systems, we look forward to working with the California Legislature and the Governor to strengthen California's SB 53," OpenAI said.
Why OpenAI changed its position
OpenAI explained its shift by pointing to the lack of comprehensive federal regulation of artificial intelligence in the United States. The company supports a model of "reverse federalism," under which individual states can jointly introduce baseline safety standards.
According to OpenAI, such state-level rules could eventually become the foundation for a nationwide standard governing advanced artificial intelligence systems in the United States.
Why this matters for legal professionals
This reversal signals a meaningful shift in how major AI companies approach state-level regulation. For lawyers advising AI developers or enterprises deploying AI systems, the practical takeaway is that California's compliance requirements may expand beyond transparency and whistleblower protections to include affirmative obligations around model monitoring and cybersecurity.
Legal teams should track SB 53 amendments closely. If OpenAI's proposal gains traction, developers of advanced models will need to document training oversight procedures and cybersecurity controls - and those requirements will likely influence contract terms, risk assessments, and liability allocation in AI-related agreements.
Your membership also unlocks:
