AI legislation in the US: Colorado and Texas open the state front

The United States has no federal AI statute, and precisely because of that, 2026 is the year of the states. Texas opened on 1 January, and the state front has only grown busier since the spring: Colorado has thoroughly overhauled its own AI Act, and California and New York now regulate the largest models — while the White House simultaneously organises a federal counterweight. For European organisations with American operations or customers, the result is a patchwork that does not resemble the AI Act.

Colorado: from the risk model to a transparency model

The original Colorado AI Act (SB24-205, signed in May 2024) was the state law closest to the EU approach: it targeted "high-risk AI systems" that make or substantially support consequential decisions on, among other things, employment, credit, housing, education, healthcare and insurance, with a duty of reasonable care against algorithmic discrimination, risk management programmes and impact assessments. But before that law ever took effect, Colorado reversed course. On 14 May 2026 Governor Polis signed SB26-189 (Automated Decision-Making Technology), which repeals and reenacts SB24-205. The new law drops precisely the three heaviest duties of the old one — risk management programmes, impact assessments and the duty of reasonable care against algorithmic discrimination — and shifts to a transparency and disclosure model: notify users when they are dealing with automated decision-making technology (ADMT), explain an adverse consequential decision within 30 days, correct inaccurate personal data on request, and provide meaningful human review. The operative duties apply from 1 January 2027. So the state that stood closest to the AI Act model deliberately stepped back from it — a telling correction against the backdrop of the federal preemption pressure (below).

Texas: the prohibition model, since 1 January 2026

The Texas Responsible Artificial Intelligence Governance Act (TRAIGA, HB 149, signed 22 June 2025) takes the opposite approach: no broad risk framework, but targeted prohibitions on AI that intentionally manipulates behaviour towards harm, intentionally discriminates, produces child sexual abuse material or unlawful deepfakes, or is used by government for social scoring. The act also creates an AI Advisory Council and a 36-month regulatory sandbox. Because of the intent requirement, the threshold for violation sits considerably higher than in Colorado or in the EU.

California and New York: a third model, aimed at the frontier

Colorado and Texas are not the whole picture. California's Transparency in Frontier Artificial Intelligence Act (SB 53), in force since 1 January 2026, adds a third and distinct model: it regulates neither everyday high-risk decisions nor specific prohibited acts, but the largest model developers themselves, requiring published safety frameworks, transparency reports and critical-incident reporting for systems trained above 10²⁶ operations. Because that tier captures the global frontier laboratories, it is the state law with the widest reach beyond US borders — and it has its own analysis in our California frontier AI law entry.

New York has since adopted the same model. The RAISE Act, finalised on 27 March 2026 and effective 1 January 2027, requires large developers of frontier models — the same 10²⁶-operations threshold and the same $500-million-revenue test as California — to maintain a published safety protocol and to report critical incidents within 72 hours, with oversight in the Department of Financial Services. Two states on exactly the same yardstick: see our analysis of the New York RAISE Act.

Washington pushes back

Meanwhile the federal stage is shifting — away from regulation and towards containing the states. As early as 11 December 2025, President Trump signed Executive Order 14365 ("Ensuring a National Policy Framework for Artificial Intelligence"), directing the Attorney General to stand up a dedicated litigation task force whose sole task is to challenge state AI laws as unconstitutional regulation of interstate commerce or as preempted, and seeking to tie federal funding to states' compliance. That task force was formally established by memorandum on 9 January 2026, but has to date not sued over any state statute. An executive order cannot by itself displace state law — preemption would require an Act of Congress. A second executive order of 2 June 2026 bet on AI innovation and security through federal criminal enforcement against AI misuse. More consequentially, on 20 March 2026 the White House published A National Policy Framework for Artificial Intelligence, a set of legislative recommendations to Congress whose seventh section urges lawmakers to preempt state AI laws that impose undue burdens with a single "minimally burdensome" national standard, declaring that "states should not be permitted to regulate AI development." It preserves states' generally applicable laws — on children, fraud and consumer protection — but takes direct aim at exactly the kind of measures Colorado and California have enacted. The Framework is not binding, but it sets the federal direction, and the tension between Washington and the states is now this year's biggest American regulatory question; this entry will be updated when it translates into adopted rules.

What this means for European organisations

Three sober observations. First: complying with the AI Act does not automatically prepare you for the American states — they have their own concepts (consequential decision, algorithmic discrimination) and their own documentation duties. Second: the models diverge sharply and are not stable — Texas rests on an intent requirement (a penal-style statute in practice), Colorado has stepped back from the AI Act-like risk model towards transparency, and California and New York target an entirely different layer, the frontier developers. There is no single "American model." Third: the direction is nonetheless clear — even without a federal statute, and despite the federal preemption pressure, AI governance in the US is becoming enforceable law, state by state.