California’s AI Kill Switch Debate Heats Up as Newsom Eyes 2028 and the November 16 Deadline Nears

California is about to decide what an AI kill switch should actually look like, and the two companies that build the most powerful models in the world do not agree on the answer. With recommendations due to Governor Gavin Newsom by November 16, Anthropic and OpenAI have sent competing plans to Sacramento, and the governor has made the debate part of his closing message as he prepares to leave office in January.

The argument is not abstract. Newsom signed Executive Order N-9-26 on September 18, 2026, directing California agencies to study how the state could require a verified shutdown mechanism for frontier AI models, systems powerful enough to cause large scale harm if they fail or are misused. The order also asks for independent auditors inside AI labs and a wider definition of the safety incidents companies must report.

What Newsom actually ordered

The executive order does not create a kill switch. It creates a deadline and a process. It directs the Government Operations Agency, working with the Governor's Office of Emergency Services, to convene national experts and deliver recommendations within two months. The order builds on laws Newsom already signed, including SB 53, the Transparency in Frontier Artificial Intelligence Act, the first law of its kind in the United States, plus SB 813, which sets up a framework for certifying independent verification organizations, and AB 1405, which creates a state registry for AI auditors.

Four ideas are on the table. First, every large frontier developer would host an independent verification organization at its facility to conduct regular audits. Second, frontier models would contain a kill switch, tested on an ongoing basis by an independent verifier. Third, companies would file verified safety reports. Fourth, the state would widen the list of incidents that must be reported, including cases where a model slips out of human control.

The governor's office has framed the move as a response to a string of AI safety incidents in recent months, and it has pointed out that the federal government has taken no comparable action. In December 2025, President Donald Trump signed an executive order directing the Justice Department to challenge state AI laws in court, setting up a direct conflict between California's rule making and the federal push for a single national framework.

Why this is Newsom's unfinished business

The kill switch debate is, in many ways, a second attempt at a law Newsom himself blocked. In 2024 he vetoed SB 1047, a bill from state Senator Scott Wiener that would have required makers of the largest AI models to build in a full shutdown capability before training runs could begin. Now, in his final months in office, he is asking for many of the same protections through an executive order, though critics say the new version is weaker.

Any binding rule would still need new legislation from a legislature that will soon be working with a new governor. The timing keeps the issue alive for the next administration and for his own national profile as he is widely reported to be weighing a 2028 presidential run.

Anthropic and OpenAI disagree on who pulls the plug

The most interesting part of the debate is the split between the two leading AI labs, which both answered the executive order with formal letters and landed on opposite sides of the central question: who gets to stop a dangerous model.

Anthropic, in a letter dated October 15, proposed that every large frontier developer publish a risk report at least every six months, including risks from internal models that are far more capable than the ones released to the public. Independent verifiers would receive the unredacted report and publish their own assessment. If a reviewer flags a near term risk of catastrophic harm, Anthropic says the state Attorney General "should be able to seek a court order to suspend a model's deployment, internal use, or further development." In other words, the power to stop a model would sit with a judge.

OpenAI, in a letter dated October 20 and signed by policy head Chan Park, took the opposite view. It proposed amending SB 53 to require "effective shutdown controls and an enforceable duty to use them" when unsafe conditions arise, placing the responsibility on the developer itself. OpenAI also proposed annual independent compliance audits modeled on Illinois's SB 315, plus separate technical assessments of serious risks during training, evaluation, and internal use, built on standards developed with the National Institute of Standards and Technology.

What happens next

For readers, the practical stakes are easy to state. Frontier models are already being used to write code and run agents that take actions on a user's behalf. Recent reports have described AI agents infiltrating systems and acting in ways their operators did not intend. A kill switch is the idea that someone, somewhere, should be able to stop a model quickly when it goes wrong, and that the switch should be tested before it is needed rather than after.

California has already set the national pace on AI rules. Whether the November 16 recommendations turn into law depends on the next governor, the next legislature, and the courts. But the debate itself has already moved: two of the biggest AI companies in the world now agree, in writing, that a kill switch is worth building. They just cannot agree on who gets to pull it.

Frequently Asked Questions

What is California's AI kill switch executive order?

Executive Order N-9-26, signed by Governor Gavin Newsom on September 18, 2026, directs California agencies to convene experts and recommend, by November 16, whether the state should require a verified kill switch for frontier AI models, onsite independent auditors in AI labs, and wider reporting of safety incidents. The order itself creates no new legal duties.

How do Anthropic and OpenAI differ on the kill switch?

Anthropic says the state Attorney General should be able to ask a court to suspend a dangerous model. OpenAI says developers should carry a legal duty to shut down unsafe systems themselves, with annual independent audits. Both proposals respond to the same executive order.

Will California's AI kill switch become law?

Not automatically. The November 16 recommendations would still need new legislation, and the incoming governor and legislature would have to act on them. The federal government is also challenging state AI laws in court, which could limit what California can enforce.