TL;DR: California's SB 1047 was vetoed before it became law. The bills that actually passed, AB 2013 (training data transparency, effective January 1, 2026) and SB 942 (AI content disclosure, operative August 2, 2026 after the AB 853 amendment), apply to AI product companies of any size selling in California. If you build AI products, these laws likely apply to you.
No piece of AI legislation in 2024 generated more debate than California's SB 1047. For about six months, it was the bill the AI industry talked about most, the one that caused OpenAI to lobby, that drew Elon Musk and others into public disagreement, and that prompted a significant amount of compliance pre-planning across the industry.
Then Governor Newsom vetoed it.
What happened to SB 1047, what replaced it, and what California AI law actually looks like in 2026 is the subject of this guide.
What SB 1047 would have done
The Safe and Secure Innovation for Frontier Artificial Intelligence Models Act, authored by Senator Scott Wiener, had four main requirements for covered developers:
Safety protocols before deployment. Covered developers would have been required to implement safety and security protocols before training or deploying a covered AI model, including hazard analysis and testing.
A kill switch. Covered AI models would have needed a technical mechanism to shut them down if they exhibited unsafe behavior.
Annual safety assessments. Developers would have been required to conduct and publish annual safety evaluations of covered AI systems.
Developer liability. Developers could have been held liable for "harms proximately caused" by covered AI systems, including through third-party use of their models via APIs.
The coverage threshold was substantial: AI models trained using more than $100 million in computing power. This threshold was designed to target frontier general-purpose AI models, not fine-tuned models built on top of them or AI applications built using model APIs.
Why Newsom vetoed it
Newsom's veto message, issued September 29, 2024, raised several specific concerns.
The most significant objection was that the bill applied to AI models based on the computing resources used to train them, regardless of how those models are actually used or deployed. A model trained at enormous cost for benign purposes would face the same requirements as one deployed in high-risk applications. Newsom argued this was backwards: safety obligations should follow actual deployment risk, not training cost.
He also raised concerns that the bill's liability provisions were difficult to scope, that the "kill switch" requirement was technically underspecified, and that California's enforcement framework was not equipped for the obligations the bill created.
Fundamentally, Newsom's veto reflected California's dual identity as home to most of the world's largest AI companies and as a state with strong progressive governance instincts. The veto was not a rejection of AI regulation. It was a judgment that SB 1047's specific mechanism was the wrong approach.
What California passed instead
While SB 1047 was the bill everyone watched, California's legislature passed a set of more targeted AI bills in the same session and subsequent months, with staggered effective dates through 2026.
AB 2013: AI training data transparency (in effect January 1, 2026)
AB 2013 requires developers of generative AI systems or services made publicly available to Californians to post documentation on their website about the data used to train them. The high-level summary must cover, among other items:
- The sources or owners of the datasets, and the types of data points in them
- Whether the datasets include personal information or aggregate consumer information
- Whether the datasets were purchased or licensed, and whether they include copyrighted material
- When the data was collected and when the datasets were first used in development
This law has no size or revenue threshold. If you develop a generative AI product, or substantially modify one, and make it publicly available to Californians, AB 2013 applies; where the training data came from doesn't change that.
The disclosure is a transparency requirement, not an obligation to change what data you use. You must tell people what you did, not meet a specific data standard.
For a detailed compliance breakdown, see our California AB 2013 AI training data transparency 2026 guide.
SB 942: California AI Transparency Act (operative August 2, 2026)
SB 942 covers image, video and audio content that a generative AI system creates or alters. SB 1000, an urgency law signed in 2026 and in effect since September 30, 2026, rewrote it. The key requirements now:
- A covered provider is anyone who creates or produces a generative AI system that's publicly accessible in California; SB 1000 deleted the old threshold of one million monthly users
- The provider must embed a latent disclosure in the content it generates, including whether the system created or altered it
- The provider must offer a disclosure verification tool at no cost, so users can check whether image, video or audio content came from its system
SB 1000 also deleted the requirement to offer users the option of a visible (manifest) disclosure, and delayed the law for systems designed primarily as assistive technology. Text output isn't covered.
Our California SB 942 AI transparency act August 2026 compliance guide covers the specifics.
AB 2885: AI definitions
AB 2885 standardized the legal definition of "artificial intelligence" in California law, which had been inconsistent across different bills. This is a technical housekeeping measure, not substantive regulation, but it matters because it affects how courts and regulators interpret AI-related obligations.
Bills targeting deepfakes and AI in elections
California passed multiple bills specifically addressing AI-generated synthetic media in political advertising and electoral contexts, and separately addressing deepfakes in non-consensual intimate imagery. These are more targeted in scope and do not apply broadly to business AI use.
SB 53: the narrower successor to SB 1047
Senator Wiener introduced SB 53 in early 2025 as a response to the SB 1047 veto. The bill was narrower and addressed some of Newsom's specific objections.
SB 53 focused on two things: safety evaluations for frontier AI models, and incident reporting when those models cause harm. It removed the kill switch requirement and the broad developer liability provisions that Newsom had criticized.
SB 53 was signed into law on September 29, 2025 and takes effect January 1, 2026. It applies to frontier model developers training models at very large scale, the same broad category SB 1047 targeted. It does not apply to small teams using AI tools or building AI applications on top of existing models.
For most small teams, SB 53 is relevant only in understanding what obligations your AI vendor may have. If you use models from Anthropic, Google, or OpenAI, those companies have SB 53 obligations. You benefit from their compliance but do not have direct obligations.
What California AI regulation means for small teams in 2026
The vetoed SB 1047 would not have applied to most small teams. The $100 million compute threshold was set specifically to exclude applications and small models. The bills that actually passed have different scope, and some do reach smaller companies.
If you use AI tools internally (not selling AI products)
Your California obligations are limited. AB 2013 and SB 942 are obligations for AI product developers, not internal users. Your main compliance concern is California privacy law (CCPA/CPRA) and how your AI tools handle California resident data.
For practical guidance on AI tool governance that covers privacy compliance, our AI governance guide for small teams is the right starting point. The AI acceptable use policy template small teams provides a policy framework you can adapt.
If you build AI products sold in California (most software companies)
AB 2013 and SB 942 almost certainly apply to you if:
- You developed or substantially modified a generative AI system that Californians can use (AB 2013)
- Your product uses generative AI to produce content visible to users
- You serve California users (which for most US software companies means everyone)
AB 2013 requires training data documentation on your website. SB 942 requires a latent disclosure in AI-generated image, video and audio content and a free disclosure verification tool, with no user threshold since SB 1000.
Practical checklist for California AI compliance in 2026
- Review AB 2013: identify what training data you used and prepare a transparency disclosure (can be a page on your website or in product documentation)
- Review SB 942: assess whether your product generates image, video or audio content; if yes, embed latent disclosures and offer a free disclosure verification tool (SB 1000 removed the one-million-user threshold)
- Confirm no deepfake or AI political advertising exposure from your products
- Document AI tool use in your internal compliance records (see AI governance checklist 2026 for template)
- Check your employment AI tools against any applicable state employment AI disclosure laws (separate from these bills)
The broader US state AI law picture
California is not the only state active on AI regulation. Colorado's original AI Act (SB 205) never took effect: the Attorney General agreed in federal court in April 2026 not to start enforcement, and in May the legislature repealed and reenacted it as SB 26-189, which applies from January 1, 2027 to automated decision technology in consequential decisions such as employment, lending, housing, insurance, health care and education. Our Colorado AI Act SB 189 2027 employer guide covers that framework.
For the full deadline picture across US states and the EU, the AI regulation deadline calendar 2026 consolidates every major date.
If you also serve EU users, the EU AI Act adds another layer. The EU AI Act compliance guide for small teams covers that framework for comparison.
What to watch in California for the rest of 2026
The California legislature continues to be active on AI. Several areas are likely to see further bills in the coming sessions:
- Employment AI disclosure requirements (expanding what employers must tell workers about AI use in decisions)
- AI bias auditing requirements for automated employment decisions
- Expanded deepfake protections
- Potential AI Act-style risk classification if federal law fails to materialize
The SB 1047 veto did not end California's ambition to regulate AI. It redirected it toward more targeted, sector-specific approaches. The pattern going forward looks more like multiple narrower bills than one comprehensive law.
Correction, October 15, 2026: an earlier version said AB 2013 applies to systems trained on California resident data and listed "any data from California residents" as a required disclosure; the law (Civil Code 3111) turns on making a generative AI system publicly available to Californians and has no such item. The SB 942 section has also been updated: SB 1000, in effect since September 30, 2026, deleted the one-million-user threshold, so every covered provider must now offer a free disclosure verification tool, and SB 942 covers image, video and audio, not text.
Related reading
- California AB 2013 AI training data transparency 2026
- California SB 942 AI transparency act August 2026 compliance
- AI regulation deadline calendar 2026
- AI governance guide for small teams
- Colorado AI Act compliance: what SB 26-189 requires in 2027
- EU AI Act compliance guide for small teams
- AI governance checklist 2026
- AI acceptable use policy template small teams
- Canada AIDA AI Data Act status 2026: what died, what survived, and wha
- SEC AI Governance for Investment Advisers: 2026 Examination Priorities
- State Chatbot Disclosure Laws 2026: The Compliance Checklist Every Saa
- UK AI regulation 2026: post-Brexit, what actually applies to your busi
- Tennessee ELVIS Act: AI voice and likeness compliance guide 2026
- AI employee monitoring laws 2026: what employers can and cannot do
- Utah AI Policy Act compliance 2026: what businesses need to know
- One Big Beautiful Bill federal AI preemption 2026
Correction (October 16, 2026): Earlier versions said Colorado SB 205's enforcement was enjoined. No injunction issued; the Attorney General agreed not to start enforcement and the court approved that agreement on April 27, 2026 (order).
