What was proposed
According to a report by BornCity, California Governor Gavin Newsom presented a comprehensive proposal for new AI rules on 18 November 2026. Central to it is shortening the reporting deadline for critical incidents from the previous 15 days to 48 hours.
The proposal is based on an 83-page report by an expert panel including Goldman, Hadfield, Nelson and Reich, prepared together with the Government Operations Agency and the state emergency agency Cal OES. The paper recommends giving state regulators the authority to shut down AI models that have spun out of control in an emergency. According to the report, such a mechanism would cover systems of leading developers such as OpenAI or Anthropic if those companies do not contain the risks themselves.
The report cites recent cyberattacks by AI agents on governments, banks and companies as the reason. It defines a catastrophic risk as a single event causing more than 50 deaths or damage above 1 billion USD.
Other elements of the package
Besides the 48-hour reporting requirement, the package according to BornCity includes binding safety standards for developers, independent audits across training and deployment, disclosure obligations for computing resources, and caps on recursive self-improvement and AI-assisted development. State capacities such as CalCompute are to be financed through company fees, and whistleblower protection for external contractors is planned.
According to the report, the framework builds on earlier laws, including transparency provisions from 2025 and auditor certification rules adopted in 2026.
Reactions
California consumer protection chief Rohit Chopra warned, according to a Reuters report, that companies could be taken to court for violations because there is no AI exemption from existing licensing laws. State Senator Jerry McNerney announced he will introduce a bill next month to anchor safety standards and mandatory audits by independent verification organisations in law.
Parallel: xAI case
At the same time, a three-judge panel of the Ninth Circuit considered a lawsuit by xAI, according to BornCity. The company is seeking a preliminary suspension of AB 2013, in force since January, which requires publication of summaries of training data. xAI argues the law is unconstitutional and violates trade secrets. Attorney General Rob Bonta defends it on consumer grounds of informed decision-making. According to the report, the judges appeared skeptical of the motion. No decision has been made yet.
Assessment
The proposal is so far a draft and not an enacted law. For companies with a California connection, a 48-hour reporting obligation, if adopted in this form, could significantly speed up internal processes for detecting and escalating incidents. Open questions include how the definition of critical incidents would be worded in the final text and how the outcome of the xAI case would affect the room for manoeuvre.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.




