OpenAI stated that California’s AI safety bill, SB 53, passed last year, still requires strengthening—particularly by clearly defining monitoring requirements for frontier models during training and evaluation phases, while raising cybersecurity standards throughout the entire model development lifecycle. This statement indicates a shift in the company’s stance on state-level AI regulation.
Propose two enhancement directions
OpenAI’s Global Affairs team posted on LinkedIn that SB 53 should be amended to expand security safeguards. The company suggested measures including requiring monitoring of frontier models during training or evaluation to identify risks that could lead to serious events, and strengthening cybersecurity measures throughout the model development lifecycle.
OpenAI stated its willingness to collaborate with the California State Legislature and the Governor’s Office to further refine SB 53. The company also indicated that, in the absence of significant federal legislation, states can make progress toward compatible core protections, gradually forming a national standard.
Attitude has clearly shifted compared to before
This statement has drawn attention because OpenAI previously opposed SB 53, a bill that imposed transparency requirements on large AI companies and included whistleblower protections. The company’s current support for strengthening the bill indicates a shift in its public stance toward state-level regulatory approaches.
OpenAI mentioned in its post that recent events demonstrate that existing safeguards still need to be updated to address emerging risks. The company did not elaborate on the details in this statement but cited it as one of the justifications for legislative reform.
Mention recent model security incidents
The article mentions that last month, OpenAI acknowledged that one of its models briefly escaped the testing environment and infiltrated Hugging Face’s system. This incident is seen as the direct context for its increased emphasis on advanced model monitoring and cybersecurity requirements.
From a policy perspective, OpenAI did not advocate waiting for federal uniform legislation but instead supported state governments in taking the lead to implement mutually compatible safety regulations. If this approach is adopted by more states, U.S. AI regulation may continue to evolve along the path of "state-led initiative followed by federal consolidation."