State-Level AI Laws in the US: A 2026 Guide for Everyone Making Content

The federal side of US AI law is a short list: the TAKE IT DOWN Act (in active enforcement), the DEFIANCE Act (stalled in the House), and a labeling bill that’s still just a proposal. State law is a different story entirely. While Congress has passed exactly one AI-specific statute that’s actually in force, individual states have been legislating at a pace that’s genuinely hard to track — as of mid-2026, roughly 109 AI-related laws are on the books across 29 states, built on top of more than 150 bills that passed in 2025 alone, with well over 1,500 more still moving through statehouses. None of this shares a single shape. Some states regulate AI in hiring decisions, some in healthcare, and a growing number regulate something that touches nearly every creator directly: what you have to disclose when the thing you’re publishing was made or altered by AI. Three states show exactly how different “compliant” can look depending on where your audience happens to be.

California: watermark it, or give people a way to check

California’s AI Transparency Act (SB 942) is the most content-specific of the three. It was originally set to take effect January 1, 2026, but a follow-up bill, AB 853, pushed the operative date to August 2, 2026, while adding new obligations for hosting platforms starting January 1, 2027. The law applies to “covered providers” — operators of publicly available generative AI systems with more than one million monthly visitors or users in California, which in practice means the major image, video, and audio generation tools rather than individual creators. Those providers have to do three things: embed a hidden, machine-readable provenance watermark in AI-generated image, video, and audio output (text-only content isn’t covered); offer a free, public detection tool that lets anyone check whether a piece of content came from their system; and give users the option to attach a visible “AI-generated” disclosure to what they produce. Civil penalties run up to $5,000 per violation per day, and the state attorney general or a prevailing private plaintiff can also recover attorney’s fees. For a working creator, SB 942 doesn’t put the disclosure burden directly on you the way an advertising law would — it puts it on the tool you’re using. But it means the platform behind your generation tool may start attaching provenance metadata to your output automatically, whether you asked for it or not, and that metadata travels with the file wherever you post it.

New York: the disclosure rule follows your audience, not your address

New York took a narrower but more aggressive approach. An amendment to the state’s General Business Law, signed into law in the winter of 2025–2026, created what’s generally described as the first state-level statute requiring disclosure specifically when advertising content uses a “synthetic performer” — a digitally created figure built with generative AI to look like a human but not modeled on any real, identifiable person. Governor Kathy Hochul’s office confirmed the disclosure requirement took effect June 9, 2026. The mechanics matter for anyone working across state lines: the law applies based on where the advertisement reaches consumers, not where the advertiser or creator is headquartered, so a brand or creator based anywhere in the country can trigger the obligation the moment a synthetic-performer ad reaches a New York viewer. The disclosure duty only attaches once the advertiser has “actual knowledge” that a synthetic performer is involved — it’s not a strict-liability trap for someone genuinely unaware their tool used one — but once that knowledge exists, the disclosure has to be conspicuous, not buried in fine print. Penalties are modest by comparison to California’s: $1,000 for a first violation, up to $5,000 for repeat violations. Small on paper, but the law is unusual in explicitly reaching outside New York’s borders, which is exactly the kind of provision that turns “one state’s rule” into “everyone’s rule” in practice.

Colorado: proof that a state’s own AI law isn’t stable, either

Colorado is worth including for a different reason: it shows how quickly a state’s flagship AI law can be rewritten out from under it. The original Colorado AI Act (SB 24-205), signed in 2024, was one of the most comprehensive state AI statutes in the country — requiring risk-management programs, impact assessments, and a duty to use reasonable care against algorithmic discrimination for “high-risk” AI systems. Its effective date was pushed back twice, and then, on April 27, 2026, a federal court enjoined enforcement entirely, in litigation brought by xAI (xAI v. Weiser) that the US Department of Justice intervened to support. Rather than fight it out in court, the Colorado legislature rewrote the law: on May 14, 2026, Governor Jared Polis signed SB 26-189, which repeals and reenacts the statute in a substantially narrower form. Gone are the risk-management program, the impact-assessment requirement, and the general anti-discrimination duty. What’s left focuses specifically on “automated decision-making technology” used in consequential decisions, with four operational duties — notify people when they’re interacting with AI, disclose adverse automated decisions within 30 days, allow correction of inaccurate personal data, and provide a path to human review. The rewritten law takes effect January 1, 2027. For creators, the direct disclosure relevance is narrower than California’s or New York’s rules, but the episode is the clearest illustration available of how unsettled this entire category of law still is — a state can pass a sweeping AI statute, watch it get enjoined by a federal court, and rewrite it into something much smaller, all within about eighteen months.

The patchwork, in practice

Line these three up and the practical problem becomes obvious. California’s rule targets the tool, triggers only above a one-million-user threshold, and centers on watermarking. New York’s rule targets the advertiser, has no user threshold, and centers on a visible disclosure tied to audience location. Colorado’s rule — at least the version taking effect in 2027 — barely touches content disclosure at all and instead governs automated decisions about people. Add in Texas, where the Responsible AI Governance Act took effect January 1, 2026 and requires government agencies and certain service providers to disclose when someone is interacting with an AI system, and Illinois, which has its own disclosure statute working through implementation, and a creator or brand publishing across state lines is left trying to satisfy several different triggers, thresholds, and disclosure formats at once — with no guarantee any two of them line up. Complicating things further, a December 2025 executive order from the White House proposed a uniform federal AI policy that would preempt state laws seen as inconsistent with it. That order is a policy directive, not a court ruling or an act of Congress, and as of this writing it hasn’t actually displaced any of the state laws above — Colorado’s rewrite came from its own legislature, not from federal preemption. Treat the preemption push as a live, unresolved fight rather than a reason to assume state rules don’t apply to you.

What this means for your workflow

You don’t need a different compliance checklist for every state your content might reach — you need one habit that satisfies the strictest rule currently in force, since a weaker rule in another state was never going to expose you to extra risk. In practice that means: default to a visible AI-generated disclosure on anything realistic enough to be mistaken for unaltered footage, especially in advertising or sponsored content; don’t assume a generation tool’s built-in watermark is doing your disclosure job for you, since California’s rule targets the provider, not your caption; and if your content or ad spend reaches a national audience, assume New York’s synthetic-performer rule applies the moment someone there sees it, regardless of where you’re based. None of the three laws above bans anything you’re doing with AI tools — they’re about what you say, and how visibly, when you use them.

State AI law is moving faster than almost any other area of regulation right now, and the details above reflect publicly available information as of mid-July 2026 — verify current requirements in the states your audience is actually in before making decisions that depend on them. This article is general information, not legal advice; for anything with real stakes, talk to a lawyer licensed in your jurisdiction.