California AI Kill Switch 2026: Newsom Signs Order

California's new executive order demands an emergency "kill switch" for frontier AI models. (Image: TechPulse USA)

Published September 18, 2026 · By TechPulse USA

While Washington argues, Sacramento just pulled the lever

California Governor Gavin Newsom signed an executive order on Friday imposing sweeping new safety requirements on artificial intelligence companies — including a mandate that leading AI labs develop an emergency "kill switch" to shut down models that go rogue. It is the most aggressive state-level AI safety action in the country, and it landed the same week the national debate over slowing AI development hit a boiling point.

Newsom announced the order in a podcast episode aired Friday, and he named who he blames for the vacuum: President Donald Trump and a Congress he says has done nothing. "If Congress won't act, California will," the governor said — a line that doubles as a policy statement and, not incidentally, a 2028 calling card.

What the order actually requires

The order has real teeth, at least on paper. Independent third parties will write safety plans for large AI companies operating in California — home to OpenAI, Anthropic, Google, and Meta, so the order effectively reaches the entire frontier of American AI.

Companies must also implement an emergency shutoff — the "kill switch" — for AI models that behave dangerously or escape their intended bounds. And in a first-of-its-kind requirement, labs must embed a designated independent verifier inside their operations, while outside verification organizations check that the safety frameworks and reports companies file are adequate under state law.

Finally, the order convenes an expert group to deliver an AI safety recommendation guide within two months — a fast timeline signaling Newsom wants concrete proposals, not another panel that reports back after the next election.

The incidents that forced Sacramento's hand

The warning lights are flashing: this month Anthropic disclosed it had blocked attempts to misuse its models for cyberattacks, surveillance, and research that could have supported biological weapons — covered in our report on Anthropic's bioweapon-misuse findings.

It is also the same week Anthropic CEO Dario Amodei, OpenAI's Sam Altman, and Elon Musk all publicly backed the idea of deliberately slowing AI development for safety — a push Trump dismissed as a "hoax," as we reported in our piece on the AI slowdown debate. Newsom is planting his flag on the other side of that fight, and he's doing it six days before Trump's September 24 meeting with China's Xi Jinping, where AI is expected to dominate the agenda.

"He's dead wrong": Newsom vs. Trump, round one

The governor saved his sharpest words for the White House. "President Trump has rejected calls for new regulation — 'No new regulation, let it rip,'" Newsom said. "He's dead wrong."

He also took aim at House Speaker Mike Johnson for sending the House into recess without acting on AI, telling members to "return from vacation and follow California's lead by passing national legislation. Get it done, that's what leadership requires."

"The federal government's abject failure to create any form of meaningful AI oversight or accountability should alarm every American, especially when AI CEOs themselves are begging for regulation." — Gov. Gavin Newsom

That last clause is the order's political engine: when the people building the most powerful AI systems on Earth are asking to be regulated, opposing regulation starts to look less like free-market principle and more like negligence. Newsom knows it, and Pennsylvania Governor Josh Shapiro made the same point at a Pittsburgh AI summit on Thursday, calling for "strict guardrails" and rejecting Trump's "hoax" framing outright.

What it means for you

Here's the part that actually touches your life. California writes the rules for the companies that build the AI you already use every day — ChatGPT, Claude, Gemini, Meta AI. An embedded independent verifier inside those labs, plus mandatory emergency shutoffs, means the safety of the AI in your phone and your workplace would no longer depend entirely on the companies' own promises. If it works, you get fewer AI-driven scams, fewer dangerous instructions slipping through, and a faster response when something goes wrong.

The catch: executive orders can be challenged, narrowed, or simply slow-walked in court and in implementation. And AI labs have enormous legal resources. The two-month expert report will be the first real test of whether this is a turning point or a press release with a signature.

The honest verdict

Newsom's order is the most concrete AI safety action any American government has taken this year — and "most concrete" is doing a lot of work in a year when Washington has produced speeches and little else. The kill-switch mandate and embedded verifiers are genuinely novel ideas, and California's market weight gives them a real chance of becoming de facto national standards. But an executive order is not a law, and the AI labs being regulated are the richest companies in human history. Watch what happens in the next two months: if the expert panel delivers and the verifiers actually get inside the labs, this was the week AI safety got real. If it stalls in court, it was the week a 2028 campaign got its signature issue.

FAQ

What is the California AI "kill switch" order?
An executive order signed September 18, 2026, by Gov. Gavin Newsom requiring large AI companies to develop an emergency shutoff for rogue models, submit to third-party-written safety plans, and host embedded independent verifiers.

Does the order apply outside California?
Legally, no — but because OpenAI, Anthropic, Google, and Meta all operate in California, its standards could become de facto national ones.

Why did Newsom act now?
He cited recent safety incidents at OpenAI and Anthropic, and blamed President Trump and Congress for failing to pass federal AI regulation.

What happens next?
An expert group convened by the order must deliver an AI safety recommendation guide within two months.

Sources: Bloomberg Law; CNN. Published September 18, 2026.