Product Widely confirmed 5 publishers 3 min read Published Sep 18, 2026, 4:25 PM PDT Updated Sep 19, 2026, 3:01 PM PDT

Newsom gives a working group two months to redraw California's frontier-model audit rules

Newsom's executive order asks an expert group he has not yet named for recommendations on onsite verifiers, independently audited risk reports and a routinely tested kill switch. California has until January 2028 to certify anyone to do the testing.

The Product Desk

Send a correction

What happened

Newsom signed an executive order on Friday telling a working group to report within two months on strengthening California's AI safety laws, including whether companies should build a kill switch for frontier models.
Newsom vetoed SB 1047 on 29 September 2024, a bill requiring developers of covered models to promptly enact a full shutdown, saying it regulated models by size and cost instead of by use.
CDT's Travis Hall has asked the governor to include advocates for civil rights and liberties in any working group set up for these processes.

Why it matters

Who has to decide decision A team shipping or fine-tuning frontier models has a planning date before it has a rule, so the defensible work is assembling audit evidence: which named person can halt a training run, and what the incident log calls a loss-of-control event.
What it costs cost On the only worked example so far, the company being tested met the tester's compute bill. Third-party evaluation sat on the developer's budget line, not the state's.
What it limits constraint California has until January 2028 to certify organisations able to test frontier models, so any requirement resting on certified testers is limited by tester capacity before it is limited by statute.
What doesn't add up contradiction Europe put the power to restrict, withdraw or recall a general-purpose model with the regulator, while California is weighing a switch built by the developer, so a company in both markets prepares for two different failure modes.

Next story · Product · Oct 4

Trump's AI accord puts three of its four layers of control inside the companies it covers

Confirmed16 publishers

The order sets one hard date, and it is for a document. A working group has two months to recommend how to strengthen California's AI safety laws ^[1]^[2]. "We're going to do this thoughtfully but with urgent velocity; the stakes are too high to wait or delay action," Newsom said in a statement ^[7]. His office has not said who is in the group ^[8].

The brief is the part a product owner can act on now. According to The Verge, the group is told to consider requiring AI companies to embed independent verification groups onsite for regular audits. It is also to consider holding transparency reports and risk assessments to the standards of independent auditors, and creating a kill switch that is routinely verified as effective. And it is to consider requiring companies to report "loss-of-control incidents" like the OpenAI attack on Hugging Face as critical safety incidents ^[5]. Three of those four items are record-keeping and access; the fourth is a build.

A first audit would ask for the name of the person who can stop a training run or pull a deployment, and the authority they do it under. Most labs already have their own means to halt rogue systems, governed by humans and in-house safety and alignment teams, and no US law requires them to, Mashable reports ^[24].

Newsom has been here before. He vetoed SB 1047 on 29 September 2024, a bill that would have obliged developers of covered models to "promptly enact a full shutdown", stopping operation and further training. He argued it regulated models by size and cost instead of by use ^[9]. This order arrives 11 days short of the two-year anniversary of that veto ^[21]. In a video on his X account, Newsom called the measure an emergency shut off, "otherwise known as an 'AI kill switch.'" He said the term "means a lot of things, depending on who you talk to" ^[4].

The people who would have to define it disagree about whether it can be defined. Mark Nitzberg, executive director of the University of California, Berkeley Center for Human-Compatible AI, told Scientific American that a universal AI kill switch is "more of an aspiration than a reality" ^[14]. Travis Hall, Director for State Engagement at the Center for Democracy and Technology, said the order "rightly prioritizes the development of better guidance regarding the definitions and requirements for AI safety regulations, particularly in regards to third party auditing" ^[16]. Hall also said CDT remains "concerned that proposals for 'kill switches' will prove ineffective in reducing risk while introducing new security risks and serious potential for government abuse" ^[17].

For teams selling into the EU, the sequencing is different. The AI Act's stop button applies to a human overseeing a deployed high-risk system and not to the model underneath, and those obligations are not yet in application ^[13].

One question is whether a rule of this shape would reach you, meaning you develop or fine-tune a model California would treat as frontier. The other is whether your incident log already separates a loss-of-control event from an ordinary outage. Answer yes to both and the next quarter is documentation. If you are in scope and the log does not make that split, the two-month report is the calendar to plan against. The auditing track is already law: the order tells a state agency to speed up the independent verifier framework and the state registry of AI auditors that Newsom signed ^[6]. SB 813's deadline for certifying organisations able to test frontier models falls roughly 14 months after those recommendations are due ^[22].

What to watch

Whether Newsom calls the special legislative session on AI he floated to Politico. That session is how the recommendations would become bills.
The names on the working group, and whether civil rights and liberties advocates get seats alongside technical experts as CDT asked.
Whether the accelerated verifier framework and auditor registry produce any certified frontier-model testers ahead of SB 813's January 2028 deadline.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence74
Adoption14
Hype gap+34
Incentives72
Confidence71

Perspective Coverage

5 publishers

Builder: Builder 31%
Operator: Operator 42%
Investor: Investor 27%

Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

[1]
Gavin Newsom signed an executive order on Friday telling a working group to report within two months on how to strengthen California's AI safety laws, Bloomberg reported; among the measures it should consider is making companies build a kill switch for frontier models.
Reported SupportedView cited source
[2]
Newsom's order directs the state to convene a group of experts that will deliver recommendations within two months on how to strengthen AI safety measures in state law.
Reported SupportedView cited source
[3]
The executive order was issued on Sept. 18 and calls for the creation of an expert-led panel tasked with "reinforcing and strengthening" the state's AI safety laws, including accelerating plans to enforce third-party safety evaluations and mandatory risk reporting.
Reported SupportedView cited source
[4]
The order directs the group to explore the possibility of an "emergency shut off" for frontier models, "otherwise known as an 'AI kill switch,'" Newsom said in a video posted to his X account, "which means a lot of things, depending on who you talk to."
Reported SupportedView cited source
[5]
Newsom wants the group to consider how the state could require AI companies to embed independent verification groups onsite for regular audits, make their transparency reports and risk assessments subject to standards of independent auditors, create a "kill switch" that is routinely verified as effective, and make sure companies are required to report "loss-of-control incidents" like the OpenAI attack on Hugging Face as critical safety incidents.
Reported SupportedView cited source
[6]
The order directs a state agency to speed up implementation of two recent laws Newsom signed, creating a framework for independent verifiers to assess AI safety, and a state registry of AI auditors.
Reported SupportedView cited source
[7]
"We're not waiting to act - we're going to speed up our work on substantial and responsible AI oversight before it's too late. We're going to do this thoughtfully but with urgent velocity; the stakes are too high to wait or delay action," Newsom said in a statement.
Reported SupportedView cited source
[8]
Newsom's office has not said who is in the working group.
Reported SupportedView cited source
[9]
Newsom vetoed SB 1047, the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act, on 29 September 2024; it would have obliged developers of covered models to be able to "promptly enact a full shutdown", stopping operation and further training, and he argued it regulated models by size and cost rather than by what they are used for.
Reported SupportedView cited source
[10]
TNW wrote of the 29 September 2024 veto that "That was almost two years ago to the day."
Reported SupportedView cited source
[11]
California's SB 813 gives the state until January 2028 to certify organisations able to test frontier models.
Reported SupportedView cited source
[12]
Since 2 August the European Commission has been able to restrict, withdraw or recall a general-purpose model from the EU market under a provision of the AI Act, and the power sits with the regulator rather than the developer.
Reported SupportedView cited source
[13]
The AI Act separately requires that a human overseeing a high-risk system can interrupt it or bring it to a halt through a stop button; that rule covers deployed systems rather than the models underneath them, and those obligations are not yet in application.
Reported SupportedView cited source
[14]
Mark Nitzberg, executive director of the University of California, Berkeley, Center for Human-Compatible AI, told Scientific American that a universal AI kill switch is "more of an aspiration than a reality."
Reported SupportedView cited source
[15]
Geoffrey Hinton told CNN he didn't think a kill switch would be an effective longterm solution for those who fear AI superintelligence.
Reported SupportedView cited source
[16]
"The California governor's executive order rightly prioritizes the development of better guidance regarding the definitions and requirements for AI safety regulations, particularly in regards to third party auditing," said Travis Hall, CDT's Director for State Engagement, in a statement to Mashable.
Reported SupportedView cited source
[17]
"We remain concerned that proposals for 'kill switches' will prove ineffective in reducing risk while introducing new security risks and serious potential for government abuse," said Travis Hall, CDT's Director for State Engagement.
Reported SupportedView cited source
[18]
Newsom is calling on Congress and Trump "to review and adopt the state's framework, or use its nation-leading regulation as a floor, not a ceiling, for the benefit and safety of all Americans."
Reported SupportedView cited source
[19]
On Wednesday, Newsom floated the possibility of calling California lawmakers in for a special legislative session on AI in an interview with Politico, alongside potential executive action.
Reported SupportedView cited source
[20]
CDT's Travis Hall said: "We call on the Governor to ensure that any working groups dedicated to these processes include a range of voices and expertise, including advocates for civil rights and liberties."
Reported SupportedView cited source
[21]
The executive order arrives 11 days short of the two-year anniversary of the SB 1047 veto.
Derived Supported
[22]
SB 813's January 2028 certification deadline falls roughly 14 months after the working group's recommendations are due.
Derived Supported
[23]
Neither California nor Europe has settled who pays for the compute for frontier-model testing; on the only worked example so far, the tester's bill was met by the company being tested.
Reported InsufficientView cited source
[24]
While not required under U.S. law, most individual AI labs have their own means to halt rogue systems, governed by humans and the company's own safety structures and alignment teams.
Reported InsufficientView cited source

Sources

5 independent publishers whose own reporting we read for this story.

engadget.com
1 article · September 19, 2026
California governor wants to implement a kill switch for frontier AI models
gizmodo.com
1 article · September 19, 2026
Newsom Executive Order Contains Eyebrow-Raising AI Kill Switch Proposal
mashable.com
1 article · September 18, 2026
Is an AI 'kill switch' a real thing? | Mashable
thenextweb.com
1 article · September 18, 2026
Newsom pitches an AI kill switch for California
theverge.com
1 article · September 18, 2026
Gavin Newsom is pushing for an AI kill switch

Clarity analysis

How the coverage leans and what the record actually supports. Read the methodology

Perspective Coverage

How much of the coverage speaks to builders, operators and investors. It measures emphasis, not whether the story is right.

Builder31%

The order hands the panel four specific things to price, and only one of them is the switch: independent verification organisations embedded onsite for periodic audits, transparency reports and risk assessments held to independent auditor standards, a kill switch whose efficacy is checked on an ongoing basis, and mandatory reporting of loss-of-control incidents. Engadget sets that beside Dario Amodei's proposal last week that labs grant third-party evaluators "ongoing, employee-like access," which is the same access question asked from the other direction. Mashable is the most useful on why the switch is hard: systems do not sit in one place, and Michael Vermeer of RAND told Scientific American that rogue agents may have to be hunted down on the open internet. For now, the order only asks a panel to study these measures; it does not yet require any lab to act.

Operator42%

Recommendations are due in two months. The machinery that would carry them is much further out. The Next Web reports that SB 813 gives California until January 2028 to certify organisations able to test frontier models, and Gizmodo, citing Politico, says the order shaves a year off implementation of the verifier framework and the auditor registry. The Next Web is also alone in noting that nobody has decided who pays for the compute a frontier evaluation consumes; on the one worked example it cites, the company being tested paid the tester's bill. In Europe the comparable stop-button duty applies to deployed high-risk systems, and it is not yet in application. The Commission's power to restrict, withdraw or recall a general-purpose model, by contrast, has been live since 2 August.

Investor27%

The Verge carries the political frame: the House is on recess until November, the midterms make Senate action unlikely, and Newsom wants Congress and Trump to treat California's rules as "a floor, not a ceiling." Gizmodo supplies the two facts with money attached. Politico's reading is that the order mainly accelerates two laws frontier AI companies already like, and Newsom's term ends on 4 January, so whoever succeeds him inherits the panel's recommendations; Xavier Becerra wrote on Threads that "We need real guardrails on companies racing toward superintelligence before anyone's proven they can control it." The Next Web adds that Newsom promoted a partnership earlier this year to put Anthropic's Claude into state agencies.

Reality

Evidence, adoption, hype gap, incentives and confidence, each read independently from the current record on the scale shown beneath it. How these are measured.

Evidence74: 0100
Five accounts, one of them quoting the text
Gizmodo prints the operative line from the order itself, which settles what is actually being asked: an assessment of "requiring the creation of a 'kill switch' for frontier models, with the efficacy of the switch verified on an ongoing basis by an independent verification organization." The Verge and Engadget agree on the four measures, and Mashable's dissent comes from named people rather than unattributed worry. The working group has no announced members, and the loss-of-control example Newsom's office reaches for, described as an OpenAI attack on Hugging Face, appears in two accounts with no description of what happened. Both gaps sit close to the center of the story rather than at its edges.
Adoption14: 0100
A report is due; nothing is in force
What exists today is a signed order asking two state agencies for recommendations, plus a year taken off the timelines of two laws that set up verifiers and an auditor registry. The obligations everyone is writing about do not bind anyone yet, and the state cannot certify a frontier-model tester until January 2028 under SB 813. Mashable reports that labs' own shutdown procedures are in-house and voluntary.
Hype gap+34: −100 understated0 aligned+100 overstated
Coverage outruns the order
Engadget's headline says the governor wants to implement a kill switch and The Verge's says he is pushing for one; the document asks a panel to study whether to require one, and Gizmodo says plainly that a governor probably cannot mandate it this way. The timing cuts the same direction: The Next Web points out that Newsom vetoed SB 1047, which would have required developers to be able to promptly enact a full shutdown, 11 days short of two years ago. Mashable and Gizmodo do most of the deflating, and both do it in the same articles whose headlines lead with the switch.
Incentives72: 0100
An order aimed at Washington
Newsom's statement opens by attacking "the federal government's abject failure" and ends asking Congress and Trump to adopt California's framework, and Gizmodo notes in passing that he is still not officially running for president. The Next Web supplies the other half: he promoted a partnership earlier this year to put Anthropic's Claude into state agencies, and both Anthropic and OpenAI sit inside whatever his panel recommends. Politico's reading, carried by Gizmodo, is that the accelerated laws are popular with frontier companies. CDT's two statements come from an organisation that opposes kill-switch mandates and is asking to be seated on the panel.
Confidence71: 0100
Firm on the document, open on the outcome
What was signed, when, and what it asks for are settled across five outlets and one direct quotation of the text. Everything downstream is open: the panel is unnamed, the recommendations are two months away, and Gizmodo notes that Newsom's term ends on 4 January, so implementation belongs to his successor.

Topics and entities

Topics

AI kill switch
AI Safety Regulation
California Tech Policy
Frontier model audits