AI Regulation in 2026: US Federal and State Laws, the EU AI Act and the UK
AI regulation in Sept. 2026: federal executive orders, state laws in Colorado, California and Texas, EU AI Act deadlines, and what small businesses must do.
Contents
As of September 25, 2026, the United States still has no comprehensive federal AI law. Federal policy runs through executive orders that favor deregulation and push back against state AI laws, while states such as Colorado, California and Texas write the rules businesses actually have to follow. If you sell to customers in Europe, the EU AI Act also applies to you: its transparency rules for chatbots and deepfakes took effect on August 2, 2026, and its “high-risk” obligations were pushed back to December 2027 and August 2028.
This guide is for people and small businesses that use AI tools rather than build them: what applies to you, what to do, and links to every official source. It reflects the texts as of Sept. 25, 2026 and is not legal advice.
Since the first version of this article (April 2025), the White House has moved to challenge state AI laws, Colorado has replaced its landmark AI Act before it ever took effect, and the EU has amended its AI Act for the first time.
AI Regulation in 2026 at a Glance
| Jurisdiction | Status (Sept. 25, 2026) | Dates that matter |
|---|---|---|
| US federal | No AI statute; executive orders, plus existing laws (FTC Act, civil rights, TAKE IT DOWN Act) | Deepfake takedown duties enforced since May 19, 2026 |
| Colorado | SB24-205 repealed and replaced by SB26-189 (automated decisions) | Jan. 1, 2027 |
| California | Frontier model safety (SB 53), AI Transparency Act, privacy rules on automated decisions | Jan. 1, 2026; Aug. 2, 2026; Jan. 1, 2027 |
| Texas | Responsible AI Governance Act (HB 149) | In effect since Jan. 1, 2026 |
| European Union | AI Act in force, applied in stages, amended by the Digital Omnibus | Aug. 2, 2026; Dec. 2, 2026; Dec. 2, 2027 |
| United Kingdom | No AI act; existing regulators and targeted offenses | Deepfake image offense since Feb. 6, 2026 |
Federal AI Policy: Executive Orders, Not a Law
Washington has not passed a general AI law. The rules come from the White House and from laws that predate generative AI.
- January 2025. Executive Order 14179 (Jan. 23, 2025) revoked the Biden administration’s 2023 AI order (EO 14110) and told agencies to remove rules seen as barriers to AI innovation.
- December 2025. Executive Order 14365 (Dec. 11, 2025), “Ensuring a National Policy Framework for Artificial Intelligence,” created a Justice Department AI Litigation Task Force to challenge state AI laws, asked the Commerce Department to list “onerous” state laws, and allowed some federal funding (including BEAD broadband money) to be conditioned on states not enforcing them. It excludes state child safety laws from the preemption it asks Congress for.
- March 2026. The White House sent Congress legislative recommendations (March 20, 2026) calling for one national standard instead of “a patchwork of conflicting state laws.” Congress had not passed such a law as of Sept. 25, 2026.
- April 2026. The Justice Department intervened (April 24, 2026) in xAI’s lawsuit against Colorado’s 2024 AI Act, arguing it violates the Equal Protection Clause.
- June 2026. A June 2, 2026 order on AI and cybersecurity sets up voluntary cooperation with frontier AI labs and states that it creates no licensing or pre-clearance requirement for AI models.
An executive order does not strike down a state law by itself: that takes a court ruling or an act of Congress. Until then, state laws apply.
Federal Laws That Already Cover AI
- Deceptive claims. The FTC Act applies to AI like any other product: exaggerated claims about what an AI tool does, or undisclosed AI use that misleads customers, can be unfair or deceptive practices.
- Fake reviews. The FTC’s rule on fake reviews and testimonials, in effect since October 2024, expressly covers AI-generated fake reviews and allows civil penalties.
- Intimate deepfakes. The TAKE IT DOWN Act (signed May 19, 2025) makes it a crime to knowingly publish non-consensual intimate images, including AI-generated ones. Since May 19, 2026, social media, messaging and image or video sharing platforms must remove reported content within 48 hours; the FTC enforces this with penalties of up to $53,088 per violation (FTC guidance, checked Sept. 25, 2026).
- Employment and credit. Anti-discrimination laws apply whether a decision is made by a person or by software.
State AI Laws to Know in 2026
| State and law | Who it covers | Key obligations | Effective |
|---|---|---|---|
| Colorado SB26-189 | Developers and deployers of automated decision-making technology (ADMT) used in consequential decisions | Technical documentation, notice to consumers, explanation after an adverse decision, human review on request | Jan. 1, 2027 |
| California SB 53 | Large frontier model developers | Publish a frontier AI safety framework, report critical safety incidents | Jan. 1, 2026 |
| California AI Transparency Act (AB 853, amended by SB 1000) | Providers of generative AI systems publicly accessible in California, with no user threshold since Sept. 30, 2026; later large online platforms, AI model hosting platforms and capture device makers | Free verification tool for AI-generated images, video and audio; latent (hidden) disclosures in that content where technically feasible; $5,000 per violation | Aug. 2, 2026 (platforms Jan. 1, 2027; devices Jan. 1, 2028) |
| California privacy rules on ADMT (CPPA) | Businesses covered by the CCPA that use ADMT for significant decisions | Pre-use notice, opt-out and access rights | Jan. 1, 2027 |
| Texas HB 149 (TRAIGA) | Developers and deployers of AI systems in Texas | Bans certain intentional harmful uses, such as intentional unlawful discrimination; civil penalties | Jan. 1, 2026 |
Dates checked Sept. 25, 2026 on the official legislature and agency pages. California AI Transparency Act rechecked Oct. 7, 2026 in the Business and Professions Code, §22757–22757.6, as amended by SB 1000, which removed the earlier threshold of over 1 million monthly visitors or users.
Colorado in detail. The 2024 Colorado AI Act (SB24-205), which required impact assessments and a duty of care against algorithmic discrimination, had already been delayed to June 30, 2026. It never took effect: SB26-189, signed May 14, 2026, repealed and re-enacted it as a lighter disclosure-based law on “consequential decisions” (education, employment, housing, lending, insurance, health care, government services). The attorney general enforces it, with a 60-day right to cure before Jan. 1, 2030, and there is no new private right of action.
If You Sell in Europe: the EU AI Act
The EU AI Act applies to companies outside the EU when the output of their AI system is used in the EU. A US business with European customers can therefore be covered, either as a “provider” (it builds an AI system or puts one on the market under its own name) or as a “deployer” (it uses an AI system in a professional context).
The Digital Omnibus on AI, Regulation (EU) 2026/1744 of July 8, 2026, was published in the EU’s Official Journal on July 24, 2026 and took effect on July 27, 2026. It is adopted law, not a proposal. The current timeline (consolidated text, Article 113, checked Sept. 25, 2026):
| Date | What applies |
|---|---|
| Feb. 2, 2025 | Banned AI practices (for example, emotion recognition in the workplace and schools, social scoring); AI literacy duty |
| Aug. 2, 2025 | Obligations for general-purpose AI model providers (OpenAI, Google, Anthropic…) |
| Aug. 2, 2026 | Transparency rules (Article 50); the rest of the Act except high-risk rules; Commission fines against model providers |
| Dec. 2, 2026 | New bans on AI that creates non-consensual sexual deepfakes or child sexual abuse material; deadline to watermark output for generative AI already on the market before Aug. 2, 2026 |
| Dec. 2, 2027 | High-risk systems in Annex III (hiring, credit scoring, education, biometrics…) |
| Aug. 2, 2028 | High-risk AI in regulated products (Annex I: toys, medical devices…) |
Transparency: Chatbots and Deepfakes
Since August 2, 2026, Article 50 requires (Commission fact page, checked Sept. 25, 2026):
- Chatbots must tell people they are talking to an AI, unless it is obvious.
- Generative AI providers must mark output in a machine-readable, detectable way (grace period to Dec. 2, 2026 for systems already on the market).
- Professional users who publish deepfakes (realistic images, audio or video of real people, places or events) must disclose that they are AI-generated. For clearly artistic, satirical or fictional work, a discreet disclosure is enough.
- AI-generated text published to inform the public on matters of public interest must be labeled, unless a human reviewed it and someone holds editorial responsibility.
The Commission published a Code of Practice on marking and labeling AI-generated content on June 10, 2026, which it and the AI Board have confirmed as adequate, plus guidelines on Article 50 and three optional EU icons for labeling. Realistic avatar videos made with tools like Synthesia or HeyGen, or a cloned voice from ElevenLabs, can fall under the deepfake rule when they depict real people.
For text, OpenAI announced on Oct. 5, 2026 that ChatGPT output will carry an invisible watermark for eligible EU users in the coming weeks, with an opt-in for API customers worldwide and a detector limited to approved researchers and expert organizations. See what ChatGPT’s EU text watermark can and can’t prove.
General-Purpose AI Models
Providers of the models behind ChatGPT, Gemini or Claude must document their models, publish a summary of training content and have a copyright policy. The GPAI Code of Practice (July 10, 2025) lists signatories including Amazon, Anthropic, Google, IBM, Microsoft, Mistral AI and OpenAI; xAI signed only the safety and security chapter, and Meta is not on the list (checked Sept. 25, 2026). This chapter creates no direct duty for users, but the training-data and copyright debate it addresses is far from settled, as the Studio Ghibli-style image controversy showed.
Penalties
| Violation | Maximum fine |
|---|---|
| Banned practices | €35 million or 7% of worldwide annual turnover |
| Most other obligations, including Article 50 transparency | €15 million or 3% |
| Misleading information to authorities | €7.5 million or 1% |
| General-purpose model providers (fined by the Commission) | €15 million or 3% |
Large companies face the higher of the two amounts; small and medium-sized businesses and startups face the lower one (Article 99). National authorities set the actual penalties.
UK: No AI Act Yet
The UK still relies on existing regulators (data protection, competition, online safety) rather than an AI statute. The King’s Speech of May 13, 2026 announced no AI bill; instead, a Regulating for Growth Bill is to set up AI regulatory sandboxes (government briefing notes). Since Feb. 6, 2026, creating or requesting a fake intimate image of an adult without consent is a criminal offense (SI 2026/31).
What Should a Small Business Do?
If you use AI tools rather than build them:
- List your AI tools and what they do: writing, images, customer chatbot, résumé screening, analytics.
- Check your claims. Don’t advertise capabilities your AI tools don’t have, and never post AI-generated reviews or testimonials.
- Handle hiring and credit decisions with care. Anti-discrimination law applies now; Colorado’s disclosure rules and California’s ADMT rules start Jan. 1, 2027. Keep a human able to review and reverse automated decisions.
- Disclose your chatbot to customers, and label realistic deepfakes. Required for EU audiences, good practice everywhere.
- Review AI-written content before publishing, whether it is written with ChatGPT or another tool, and keep editorial responsibility.
- Protect personal data. Don’t paste customer or employee data into a tool without checking its data-use settings and your privacy obligations.
- If you have EU customers or staff, train the people who use AI tools (the EU’s AI literacy duty) and check that no use falls under the EU’s banned practices, such as emotion recognition of employees.
For a specific product or use case, the EU’s AI Act Service Desk and a lawyer in the relevant state or country are the right places to go.
FAQ
Is there a federal AI law in the United States?
No. As of Sept. 25, 2026, there is no comprehensive federal AI statute. Federal policy comes from executive orders, and existing laws (FTC Act, anti-discrimination law, the TAKE IT DOWN Act) apply to AI. The White House has asked Congress for a national law that would preempt most state AI laws, but none has passed.
Has the federal government blocked state AI laws?
Not by itself. Executive Order 14365 directs the Justice Department to challenge state laws in court and ties some federal funding to them, but only courts or Congress can invalidate a state law. State laws in Colorado, California and Texas remain on the books.
When does the Colorado AI Act take effect?
The original 2024 law never took effect. Its replacement, SB26-189, signed May 14, 2026, applies from January 1, 2027, and focuses on notices and explanations for automated decisions in areas like hiring, lending and housing.
Does the EU AI Act apply to US companies?
Yes, when the output of their AI system is used in the EU, for example a chatbot serving European customers or AI-generated content published to a European audience. Purely personal, non-professional use is not covered.
Was the EU AI Act delayed?
Partly. The Digital Omnibus (in force since July 27, 2026) moved high-risk obligations to Dec. 2, 2027 and Aug. 2, 2028. Bans, general-purpose AI rules and the Article 50 transparency rules were not delayed.
Do I have to label AI-generated content in the US?
There is no general federal labeling rule. California’s AI Transparency Act puts disclosure duties on providers of generative AI systems publicly accessible in California, not on their users; since SB 1000 took effect on Sept. 30, 2026, it applies regardless of a provider’s number of users (checked Oct. 7, 2026). The FTC can act against content that deceives consumers. If your audience includes people in the EU, deepfakes and unreviewed AI news text must be labeled.