OpenAI wants California to strengthen its newly passed AI safety law

OpenAI is urging California to bolster its newly enacted AI safety law, SB 53, which it played a role in passing earlier this year. The company argues the existing legislation could benefit from increased provisions to monitor and secure AI models during training and evaluation. OpenAI specifically calls for measures that would detect potential breaches of third-party security controls and unauthorized access to confidential data.

In addition to these training safeguards, the tech giant is advocating for expanded cybersecurity protocols throughout the entire model development cycle. These measures aim to thwart AI systems from circumventing their internal safeguards.

SB 53, which Governor Gavin Newsom signed into law in September 2025, already represents a substantial piece of legislation. It mandates large frontier AI developers to publish safety frameworks, establishes a formal process for reporting critical safety incidents to the state's Office of Emergency Services, protects whistleblowers, and empowers the Attorney General to impose civil penalties on non-compliant companies.

The law also created CalCompute, a public computing consortium designed to foster safety and equity research, and requires the state to revisit and update the law annually as the underlying technology evolves.

Newsom touted SB 53 as California's ability to protect the public while maintaining the state's robust AI industry. The law has been praised as a temporary solution to the absence of comprehensive federal AI policy in Washington. OpenAI sees its proposed amendments as a proactive approach, rather than a reactive measure, to address potential issues before they escalate.

The company emphasizes this isn't about crafting rules for a specific incident, but rather about establishing a framework that can identify problems earlier and ensure industry-wide learning. OpenAI's push for these changes not only keeps them involved in shaping safety standards but also underscores their commitment to enhancing AI safety and reliability.