California signed thirteen AI laws in a single afternoon

Thirteen bills, one afternoon

On September 30, 2026, Gavin Newsom signed thirteen laws on artificial intelligence. Not thirteen statements of intent: thirteen chaptered statutes now sitting in California's code, each with its own deadlines and fines.

One example shows the scale. A California employer will no longer be able to base a disciplinary action or termination on an automated system's decision alone. When the machine is the main driver, a human has to corroborate the call with supporting evidence, and the employee gets a written notice explaining what information about them was used. That's SB 947, in effect July 1, 2027.

The fine is $500 per violation. The number itself isn't the point. The fact that there's a number is.

Why it all lands on the same day

The date isn't a stunt. Article IV, section 10, paragraph (b)(2) of the California Constitution says a bill passed before September 1 of the second year of the session, and still sitting on the Governor's desk, automatically becomes law unless it's vetoed by September 30.

September 30 works like a cash register at closing time. Whatever's still on the desk goes one way or the other, and the Governor signs in bulk.

That explains the day's format. It doesn't explain what's in it.

At work, AI becomes something you have to report

Two bills touch layoffs directly. SB 947 bars a fully automated firing decision, as above. SB 951 goes further and rewrites the state's mass-layoff law.

When a layoff is driven mainly by AI or automation, the notice must now carry the line "This notice is for a technology displacement," the number and classification of eliminated positions, the functions being automated, and the specific system involved. The state's employment agency publishes these notices and rolls them into a quarterly tally.

In other words, California is building itself a counter. Until now, "AI is wiping out jobs" was a sentence. From here on, it's a checkbox on a form, and somebody adds up the boxes.

Two more bills deal with surveillance. AB 1331 bars an employer from running a surveillance tool in workplace restrooms, and gives employees the right to leave their device outside before going in. AB 1883 bars devices that collect neural data or claim to predict an employee's emotional state.

Where the machine no longer gets to decide

SB 574 targets lawyers and arbitrators. A lawyer can't hand off legal work to generative AI. They have to verify every case citation and legal reference the machine produces, fix what's wrong, and disclose to the court that AI was used. An arbitrator can't delegate the ruling itself.

The bill names its own cause, since it requires checking every single citation of case law and statute. California is turning this into a professional duty rather than a case-by-case penalty.

On healthcare, AB 1979 requires clinicians to keep their own independent judgment, bars AI from performing a licensed clinical function on its own, and requires disclosure whenever AI drafted a clinical message sent to a patient. Health chatbots become care providers, which means they're now bound by patient confidentiality. SB 503 adds a requirement to hunt for bias in clinical decision-support tools.

AB 2713 bars large platforms from knowingly stripping out standards-compliant provenance data, the markers that say where an image came from. SB 1111 extends likeness rights to digital replicas of voice and face. And SB 1159 simply excludes AI, agents, and robots from the definition of "person" in the state's transparency laws: a machine can't approach a government agency pretending to be a constituent.

Two bills stand apart from the rest. AB 1864 requires makers of benchtop nucleic-acid synthesis devices, and sequence providers, to follow a customer-screening framework published in 2024, and to publish a compliance attestation, under a $5,000-a-day penalty. AB 2392 sets procurement standards and mandatory generative-AI training for public universities, requiring that the systems they adopt be guarded against harmful content and against models that are excessively sycophantic. The same afternoon, then, produced a biosecurity rule and a rule against chatbots that are too agreeable.

California isn't satisfied with its own laws

That same day, Newsom sent the Assembly a signing message worth reading. In it, he writes that the two workplace-surveillance bills leave work undone, and asks lawmakers to take them back up.

His examples are specific: AB 1331 risks getting in the way of systems employees use to unlock secured spaces, and AB 1883 lacks definitions, which will create confusion over which tools are actually covered.

A governor who signs a bill while pointing out the holes in it is rare, and more honest than most press releases. It's also a reminder of the real scale of the event: the automated-termination rule doesn't take effect until July 1, 2027, the content-provenance rules kick in on January 1, 2027, the university procurement standards aren't expected until 2028, and a California law is still, in the end, a California law.

The day before, in Washington

The day before, on September 29, Donald Trump signed Executive Order 14434, titled "Inaugurating the Era of Super Intelligence" and published in the Federal Register on October 2. According to the White House fact sheet, it directs federal agencies to use the terms "Super Intelligence" and "SI" in their communications, and to stop recognizing the terms "Artificial Intelligence" and "AI." The new name, it says, better captures capabilities that free up human creativity instead of imitating it.

That same day, six industry leaders, Sundar Pichai, Dario Amodei, Mark Zuckerberg, Greg Brockman, Elon Musk, and Jensen Huang, signed a frontier-model safety agreement with the White House. A voluntary commitment, four announced layers of internal controls and external audits, and no law behind any of it. It isn't the first time a federal AI order has been worked out in a room with these same people: we already reported on how three phone calls buried one such order.

One last signature from Newsom remains, from that same September 30. A one-page order, a single paragraph commanding: every agency under his authority will keep saying "Artificial Intelligence" and "AI," "notwithstanding any rebranded or different terminology used by the federal government." The text specifies that it creates no enforceable right, and the preamble before it notes that renaming a thing distracts no one from the inaction on its risks.

Thirteen laws that compel. And one order that does nothing but hold onto the word.