AI law and the newsroom, October 1: California signs the rest of its AI batch, provenance marks get protection, and Google goes to the EU court

This issue covers 1 October 2026: four items from two jurisdictions. California's governor signed the AI bills still sitting on his desk when the last issue went out, and that batch decides what audiences are told about synthetic content, who counts as a person in the state's open-government laws, and how employers may use automated decisions. Google asked an EU court to suspend an order that would hand its search data to AI chatbots.

Three questions run through the items below, and this digest keeps asking them. What may AI companies do with your work? What must you tell your audience when you use AI? Which office, and which document, will ask you about it?

Gov. Gavin Newsom signed the bills left on his desk at his 30 September deadline, among them the three this digest has been tracking: SB 1000 and AB 2713, which amend the California AI Transparency Act, and SB 1159. 1

SB 1000 took effect when he signed it, because the bill declares itself an urgency statute. It deletes the one-million-monthly-user threshold from the Act's "covered provider", so the law now reaches anyone who creates, codes or otherwise produces a generative AI system that is publicly accessible in California. It renames the free "AI detection tool" a "disclosure verification tool", and the statute says what that tool must do: take an upload or a web address, work through an application programming interface, output the system provenance data it finds, and hold back personal information unless the person it belongs to consents. Providers that make or alter image, video or audio must, so far as is technically feasible, embed a latent disclosure saying whether the system created or altered the content, and make that disclosure permanent or extraordinarily difficult to remove. 2

Two details matter at a desk. The tool is free and can be called without visiting the provider's website, which makes it a check on a user-submitted photo, video or audio clip. A system built mainly to serve as assistive technology stays outside the Act until 1 January 2029, and a provider that falsely claims that status faces $50,000 per violation against $5,000 for everything else, with each day counted as a separate violation. 2

AB 2713 is the platform side, and it starts on 1 January 2027. A large online platform — a public-facing social media, file-sharing or mass messaging platform, or a stand-alone search engine, above two million unique monthly users in the preceding 12 months — must detect provenance data on the content it distributes, tell users whether that content came from a generative AI system or from a capture device, name the system or the device, say whether a digital signature is attached, and let a user inspect the system provenance data by display, link or download in a form that is hard to attach to other content. The platform must keep system provenance data and digital signatures intact where that is technically feasible, and data that is not interoperable with widely adopted standards falls outside both duties. 23

SB 1159 does one thing across seven statutes. "Person", "interested person", "participant", "member of the public" and similar terms in the California Public Records Act, the Legislative Open Records Act, the Bagley-Keene Open Meeting Act, the Ralph M. Brown Act, the Administrative Procedure Act, the California Coastal Act and the California Environmental Quality Act mean natural persons, and do not take in AI systems, autonomous agents or robots. A separate section makes it unlawful to use AI knowingly to represent that a natural person appeared before, submitted information to, or otherwise engaged with a government agency. 4

Agencies get a way to hold that line: a public agency, a state body, the Office of Administrative Hearings and the Legislature's rules committees may run a disclosure verification tool built to the same standard SB 1000 sets for providers. A natural person may still use AI to help them deal with an agency, provided the volume and frequency stay reasonably consistent with ordinary participation. The Legislature's findings name what prompted the bill — records requests arriving in the thousands, comments generated in bulk on proposed regulations, automated petitions — and point to AI-generated planning objections in the United Kingdom as the warning. 4

SB 1159 carries no urgency clause, so it takes effect on 1 January 2027. 4 For a newsroom its weight sits in local coverage: public comment, records requests, and proceedings under the environmental and coastal statutes now belong to people.

SB 947 starts on 1 July 2027 and puts automated decision systems into the Labor Code. An employer may use one only within limits: it must not break labour, safety, employment or civil-rights law, infer an employee's protected status, or punish a worker for exercising legal rights. For a disciplinary or termination decision the employer has to keep a person in the loop. It may not rest on the system's output alone, and where that output carries the decision, a human must corroborate it against the data behind the output or other supporting material. An output that nobody can corroborate, or that the reviewer finds inaccurate, incomplete or misleading, stays out of the decision. The employee can ask for a description of their own data that the system used, and must be told in writing what part the system played. The Labor Commissioner enforces this part at $500 per violation, and a union contract with an explicit waiver can displace it. 5

SB 951 changes what a mass-layoff notice says. Where an AI system or other automation takes over jobs and causes a layoff, relocation or termination in whole or in substantial part, the 60-day notice under California's WARN Act must carry "This notice is for a technology displacement" at the top, the number, classification and location of the affected jobs, the job functions the technology will perform, and the category or type of system involved. The Employment Development Department must publish a summary of those notices and a quarterly statewide count, and report to the Legislature by 1 January 2028 on how AI affects hiring. 6

SB 574 governs the lawyers a newsroom hires. An attorney must keep the practice of law in human hands, so generative AI can assist the attorney and cannot take the work over. The attorney may not put confidential or non-public personal information into a system open to anyone beyond the attorney and others bound by confidentiality, must take reasonable steps to check the output — every case and statutory citation included — and must correct anything hallucinated. A document filed with a court must say that generative AI helped prepare it, and no paper filed in a California court may carry a citation that the responsible attorney has not personally verified, including one the AI supplied. An arbitrator may not hand any part of the decision to that kind of tool. 7

SB 951 and SB 574 are regular-session bills with no urgency clause, so they take effect on 1 January 2027. 67 The same signing covered bills outside this digest's three questions: digital replicas (SB 1111), clinical AI (AB 1979 and SB 503), workplace surveillance (AB 1331 and AB 1883), higher-education procurement and training (AB 2392), and gene-synthesis safeguards (AB 1864). 1 AB 1609, on customer-service chatbots, was approved on 28 September and chaptered as Chapter 733 of 2026. 8

Google went to the General Court of the European Union on 29 September to annul two binding specification decisions the European Commission adopted on 16 July, and applied on 30 September for interim measures to suspend one of them. 9

The two decisions tell Google how to meet its obligations under the Digital Markets Act: give rival AI services equal access to features on Android, where Google's own Gemini sits, and share anonymised Google Search data with third parties, among them AI chatbots that offer search. Google says the search-data order puts the privacy of European users at serious risk. The Commission says the decisions already account for integrity and security features and for the protection of end users' personal data. 910

The order settles which AI assistants may draw on the search data that AI answers are built from. The Commission has a separate, still-running investigation into whether Google's AI Overviews and AI Mode are built on publishers' content without fair compensation or a workable opt-out; it opened in December 2025. The July decisions stay binding while the General Court hears the case, so nothing decided in the past two days changes what an AI company may do with a news publisher's material. Google's filing is dated 29 September, the day before this issue's window opens, and it earns its place here by putting the next date on that file in front of readers who follow it. 911

The nearest dated item is 2 December 2026, when the EU's prohibitions on AI systems that generate non-consensual intimate material and child sexual abuse material start to apply. Then 1 January 2027 carries four California obligations from this batch — AB 2713, SB 1159, SB 951 and SB 574 — with 1 July 2027 following for SB 947. Behind those sit 2 December 2027 and 2 August 2028 for the EU's high-risk rules.

Two things in this file have no date yet. The General Court has not said when it will rule on Google's request to suspend the search-data order. New York's Stealth Crawler Prohibition Act, passed in June to make unidentified AI crawlers identify themselves to news sites, has sat on Gov. Kathy Hochul's desk since then, and a law-firm alert published on 11 June expected no action before the state's election on 7 November. 12