If you read only one headline about AI regulation 2026 this year, it was probably “EU delays AI Act.” It is half true, and the half that is not true is the part that trips people up. The EU did postpone its high-risk AI rules, to December 2, 2027 and August 2, 2028. But the rest of the Act did not wait: the bans and AI literacy duties have applied since February 2025, the rules for general-purpose AI model providers since August 2025, and the transparency duties and the Commission’s enforcement powers since August 2, 2026. More deadlines land on December 2, 2026.

Meanwhile, the United States still has no federal AI statute, and a fight between a federal preemption push and state laws is under way. China regulates AI through targeted rules, including mandatory labels on AI-generated content. South Korea’s AI Basic Act has applied since January 22, 2026, though fines are on hold for at least a year. Japan, India and Singapore rely largely on guidance and promotion.

This guide maps all of it in one place: what applies today, what was delayed, who is covered, and the dates to put in your calendar. It ends with a practical checklist by role, because the right question is not “what do the laws say?” but “which of them apply to me?”

How this guide was researched

We used law-firm and compliance trackers (DLA Piper, Gibson Dunn, Latham & Watkins, and specialist AI-policy sites), a regulation tracker for the EU, a university tracker for U.S. states, and Asian policy trackers. For the EU, several independent sources agreed on the dates after the Digital Omnibus. We did not read the official legal texts in full.

Three limits matter. Most sources are secondary, and some are vendor or consultancy sites, so we label them and flag where they disagree (for example, on Colorado’s law and on Japan’s dates). Rules change quickly, especially in the U.S. Nothing here is legal advice. Where we give a fine or a date, check the official source.

AI Regulation 2026: The map in one table

Status as of November 13, 2026, as reported by the sources named at the end.

Jurisdiction Main instrument(s) Status Style Penalties (as reported)
European Union AI Act (Regulation 2024/1689), amended by the Digital Omnibus (Regulation 2026/1744) Bans, AI literacy, GPAI duties, transparency and Commission enforcement are in effect; high-risk delayed to Dec 2, 2027 and Aug 2, 2028 Comprehensive, risk-based The Act sets tiered maximum fines; verify the amounts in the text
United States No federal AI statute; Executive Order 14365; state laws (California, Texas, Illinois, Colorado and others) State laws in force; federal preemption pursued, not achieved Patchwork Varies by state
China Targeted rules, including AI content labeling measures and standard GB 45438-2025; amended Cybersecurity Law with an AI clause In force Targeted, content- and platform-focused Cybersecurity Law fines up to RMB 10 million (secondary source)
South Korea AI Basic Act and Enforcement Decree In force Jan 22, 2026; fines paused for at least a year Framework law with high-impact AI duties Up to KRW 30 million (about US$21,000)
Japan AI Promotion Act In force (fully from Sept 1, 2025) Promotion, no penalties None
India AI Governance Guidelines (non-binding); IT Rules amendment on synthetic content Guidelines Nov 5, 2025; labeling rules Feb 20, 2026 Guidance plus platform duties Under the IT Rules for platforms
Singapore Model AI Governance Frameworks (voluntary), including one for agentic AI; Personal Data Protection Act In force as guidance Voluntary None specific to AI
Taiwan AI Basic Act In force Jan 14, 2026 Mainly directs government agencies Not verified
Vietnam AI Law In force Mar 1, 2026 Comprehensive Not verified

The European Union: the AI Act after the Digital Omnibus

The European Union: the AI Act after the Digital Omnibus
The European Union: the AI Act after the Digital Omnibus

The background. The AI Act entered into force on August 1, 2024 and phases in obligations over several years. In November 2025, the Commission proposed a “Digital Omnibus” that would defer the high-risk rules.

Parliament approved it on June 16, 2026 (423 votes to 57, with 174 abstentions), the Council adopted it on June 29, and Regulation (EU) 2026/1744 was published in the Official Journal on July 24 and entered into force on July 27, 2026, before the original August 2 date, according to several trackers. A consequence several sources note: the high-risk provisions of the original Act never applied in their original form.

What applies now, and what was delayed: 

Obligation Original date Current status
Prohibited AI practices (Article 5) and AI literacy duties Feb 2, 2025 In force
General-purpose AI (GPAI) model provider obligations (Articles 51 to 56) Aug 2, 2025 In force; unchanged by the Omnibus
Commission enforcement powers over GPAI providers Aug 2, 2026 In force since Aug 2, 2026
Transparency duties (Article 50): telling users they are interacting with AI, labeling deepfakes and AI-generated content Aug 2, 2026 In force (with a later date for one marking duty; see below)
Machine-readable marking of AI-generated content for generative systems already on the market before Aug 2, 2026 Aug 2, 2026 Dec 2, 2026
Two new bans, on non-consensual intimate imagery and child sexual abuse material generated by AI Not in the original schedule Dec 2, 2026
National AI regulatory sandboxes operational Aug 2, 2026 Aug 2, 2027
High-risk AI, stand-alone systems (Annex III: hiring, credit, education, law enforcement, border control and similar) Aug 2, 2026 Dec 2, 2027
High-risk AI embedded in regulated products (Annex I) Aug 2, 2027 Aug 2, 2028
Legacy high-risk systems used by public authorities Aug 2, 2030 Unchanged

The December trap. One analysis warns that a company that read “AI Act delayed” and paused its compliance work is already behind on some obligations and close to the line on others, with December 2, 2026 the next date to watch, for the two new bans and for machine-readable marking on generative AI systems that were already on the market.

Who is covered. The Act applies to companies outside the EU that place AI on the EU market or whose AI output is used in the EU, according to summaries, and the GPAI duties bind model providers, not the companies that merely call their APIs. If you build on a model provider’s API, the provider carries the GPAI model obligations, while you may have transparency and, later, high-risk duties depending on your use. We summarize this from secondary sources, so confirm your own position.

What Article 50 means in practice. Where people interact with an AI system, they must generally be told, and certain AI-generated or manipulated content, such as deepfakes, must be labeled. This is directly relevant to anyone publishing AI-generated images, audio or video.

What was not weakened. Law-firm commentary stresses that the Omnibus is a deferral, not a dismantling: the Act’s risk-based architecture, governance structure and core obligations remain.

The United States: no federal law, a state patchwork and a preemption fight

The United States: no federal law, a state patchwork and a preemption fight
The United States: no federal law, a state patchwork and a preemption fight

No federal AI statute. As of November 4, 2026, one legal-tracker found no federal law setting general rules for private AI systems or generally preempting state AI laws. The federal government has acted through executive orders and policy statements.

The federal push. According to trackers, Executive Order 14365 (December 11, 2025) directed a Department of Justice AI Litigation Task Force, established on January 9, 2026, to challenge state AI laws, a Commerce Department evaluation of “onerous” state laws, and a legislative recommendation for a preemptive federal framework. A March 2026 White House framework urged Congress to pass such a standard, and a June 2026 executive order on advanced AI innovation and security followed, per one briefing.

The trackers agree on the key point: preemption has been pursued, not achieved. An executive order does not void a state statute, and Congress has twice declined to pass a moratorium, according to one source.

State laws in force. The state map is active, with the following in force or scheduled, as reported by trackers:

State or city Law Status What it targets
California SB 53 (Transparency in Frontier AI Act) In force Jan 1, 2026 Large frontier developers must publish safety frameworks and report catastrophic-risk incidents; it binds your model provider more than you
California Automated decision-making rules and other laws; 20+ AI laws took effect Jan 1, 2026 In force Employment and consumer decisions, training-data disclosure (AB 2013) and more
Texas TRAIGA (HB 149) In force Jan 1, 2026 Bans AI developed or deployed with intent to cause listed harms, plus government and health-care disclosure; the attorney general enforces, with a cure period
Illinois HB 3773 In force Jan 1, 2026 AI in employment
New York City Local Law 144 In force Bias audits for automated hiring tools
Utah Generative AI disclosure law In force Disclosure
Colorado Original AI Act (SB 24-205) rewritten as SB 26-189, signed May 14, 2026 Sources disagree: two trackers say the replacement takes effect Jan 1, 2027; another says the original became operative June 30, 2026 and that a court order paused enforcement after the federal government joined a challenge by xAI Automated decision-making
New York (RAISE) and Illinois (frontier) Frontier-model laws Queued for 2027 Frontier developers

Litigation. One legal-tracker says the United States joined xAI’s challenge to Colorado’s law, and that an April 27 stipulated order paused Colorado enforcement. Another summary says the Task Force had not filed any lawsuits as of mid-2026. These accounts are not necessarily inconsistent (one is about intervention in a private case), but check the court record before relying on either.

United States joined xAI's challenge
United States joined xAI’s challenge

The next date. California’s Governor signed Executive Order N-9-26 on September 18, 2026, directing state agencies to deliver recommendations by November 16, 2026 on amending the state’s AI safety laws.

What this means. For a U.S. business, compliance is a two-front problem: obey the state laws that apply to your activity and customers, and watch for federal preemption efforts that could change them. Because state laws apply based on where people are, a company selling nationally may face several at once.

China: targeted rules and mandatory labels

China: targeted rules and mandatory labels
China: targeted rules and mandatory labels

China has no single AI Act. It regulates AI through a dense set of targeted, binding rules issued under other laws, according to several trackers.

• Mandatory labeling of AI-generated content. The Measures for Labeling AI-Generated Synthetic Content and the national standard GB 45438-2025 took effect on September 1, 2025. According to Latham & Watkins’ summary, providers must add explicit, user-visible labels to AI-generated text, images, audio and video where the content could mislead the public, and embed implicit, machine-readable labels (metadata with provider details and reference numbers) in all AI-generated synthetic content. Content-sharing platforms must verify incoming metadata, add labels where a user declares content as AI-generated, and flag suspected AI content when identifiers are missing.

• An AI clause in the Cybersecurity Law. The amended Cybersecurity Law took effect on January 1, 2026 with an AI governance clause and tiered penalties, with fines of up to RMB 10 million, according to one tracker. We did not verify that figure against the legal text.

• Filings and content rules. Commentary says generative AI services serving people in China face filing and security-assessment requirements and content rules, which can create a heavy operational burden for foreign providers. We did not verify the details.
What this means. If your product or content reaches users in China, expect labeling, platform and content-governance duties, and take local advice. The focus is on content and services, not on a single risk-tier system like the EU’s.

South Korea: a framework law in its grace period

South Korea’s Framework Act on the Development of Artificial Intelligence and the Creation of a Foundation for Trust, known as the AI Basic Act, and its Enforcement Decree took effect on January 22, 2026. Trackers describe it as the first comprehensive AI framework law in Asia and the first after the EU’s.

• High-impact AI. Systems in fields such as health, energy, hiring and lending face duties on notice, risk management, human oversight, explanation and user protection.

• Generative AI labeling. Operators that provide generative AI, or products using it, must label outputs to indicate they were AI-generated (Article 31, per one tracker).

• Reach beyond Korea. Larger foreign providers must appoint a domestic representative, and additional duties apply to models above a compute threshold of 10^26 FLOPs, according to a compliance summary. We could not verify the thresholds.

• Modest fines, and a grace period. The maximum administrative fine is KRW 30 million, about US$21,000, covering three failures only, and the Ministry of Science and ICT has indicated a grace period of at least a year before fines are imposed, with 2026 effectively a pilot period.

• A national plan. Korea’s AI Action Plan 2026 to 2028 was finalized on February 25, 2026, per one policy tracker.

The lesson is that Korea’s law is real and in force, but its penalty regime is deliberately gentle for now. The grace period ends no earlier than January 2027.

Japan, India, Singapore, Taiwan and Vietnam

Japan: promotion over prohibition. The AI Promotion Act, enacted in 2025 and described as fully in force from September 1, 2025 (one source gives June 4, 2025 for entry into force), creates an AI Strategy Headquarters under the prime minister and imposes cooperation duties but no penalties. A first AI Basic Plan was adopted in December 2025 according to two trackers, and another says the current plan was adopted on July 14, 2026, so check which is current.

India: guidance plus platform rules. Non-binding AI Governance Guidelines were issued on November 5, 2025, and mandatory AI-content labeling rules for platforms under the IT Rules began on February 20, 2026, according to trackers. India also runs the IndiaAI Mission.

Singapore: voluntary frameworks. Singapore uses its Model AI Governance Framework, including guidance on generative AI and, since January 22, 2026, a framework for agentic AI, alongside the Personal Data Protection Act and the AI Verify toolkit. One tracker notes it bans candidate deepfakes during elections.

Taiwan and Vietnam. Taiwan’s AI Basic Act took effect on January 14, 2026 and mainly directs government agencies, according to a tracker. Two sources say Vietnam’s AI Law has been in force since March 1, 2026, and one calls it comprehensive. We did not verify details for either.

The United Kingdom. One comparison says the UK still has no AI bill and relies on principles and existing law.

Four regulatory models

Model Examples Key feature Trade-off
Comprehensive, risk-based law EU, South Korea, Vietnam One statute, with tiers of obligation Clear rules, but heavy compliance and frequent delay
Targeted rules China Binding rules on specific uses such as labeling and generative services Fast to adapt, but fragmented and demanding for foreign firms
Promotion and soft law Japan, India (guidance), Singapore, UK, Taiwan Strategy, principles and voluntary frameworks Flexible, but little legal certainty or enforcement
Patchwork United States State laws, executive orders and sector rules, with no federal statute Innovation-friendly at the top, but conflicting rules and litigation

Common threads

Across very different systems, the same ideas keep recurring:

• Transparency about AI content. The EU (Article 50), China, South Korea and India all require labeling or disclosure of AI-generated content in some form.

• Extra duties for high-impact uses. Hiring, credit, health and public services are targeted by the EU, South Korea and U.S. states.

• Frontier-model governance. The EU’s GPAI rules, California’s SB 53 and Korea’s compute threshold all aim at the most capable models.

• Reach beyond borders. The EU, Korea and China all apply to foreign providers serving their users.

• Delay and grace periods. The EU deferred high-risk rules, Korea paused fines for at least a year and Colorado rewrote its law before it applied. Compliance dates move, so build processes that do not depend on one date.

Which rules apply to you? A role-based guide

This is our editorial guidance, built from the sources above. It is not legal advice, and your facts may differ.

If you are… Rules likely to matter Practical starting steps
A blogger, creator or publisher using AI-generated images, audio or video EU Article 50 labeling of deepfakes and AI-generated content; labeling rules in China, South Korea and India if you serve those users or platforms Label AI-generated media clearly, disclose AI assistance in your editorial policy, keep records of what you generated, and avoid realistic fakes of real people
A startup building on third-party AI model APIs Transparency duties (telling users they are interacting with AI); later, high-risk duties if your use case is listed; your provider carries the GPAI model duties Inventory your AI uses, review vendor terms, add AI-interaction notices, and classify any hiring, credit, education or health use as potentially high-risk
A company using AI in hiring, credit or similar decisions NYC Local Law 144, Illinois HB 3773, California automated-decision rules, South Korea’s high-impact duties, EU Annex III from December 2, 2027 Run bias audits where required, keep a human in the loop, document decisions and give people notice and explanations
A provider of general-purpose or frontier models EU GPAI duties and Commission enforcement since August 2, 2026; California SB 53; South Korea’s domestic-representative and compute rules Maintain technical documentation, publish required frameworks and incident reporting, and appoint local representatives where required
A Web3 project using AI agents or generated media AI-interaction disclosure and deepfake labeling; plus financial and data-protection rules that vary by country Disclose when users are dealing with an AI agent, label generated content, keep personal data off-chain and take legal advice
An employee using AI tools at work EU AI literacy duties on organizations; data-protection rules wherever you work Do not paste personal or confidential data into tools without approval, and follow your employer’s policy

A ten-step starter checklist for any organization:

  1.  List every AI system you build, buy or use, including free tools staff use.
  2.  Note where each is used and who is affected, and in which countries.
  3.  Classify each use by risk, with extra care for hiring, credit, education, health and public services.
  4.  Add clear notices where people interact with AI.
  5.  Label AI-generated or manipulated content, visibly and, where possible, with metadata.
  6.  Review vendor contracts for data use, security and regulatory responsibilities.
  7.  Check data-protection obligations for any personal data involved.
  8.  Keep a human in the loop for significant decisions, and document them.
  9.  Train staff, which the EU’s AI literacy duty makes a legal requirement for many organizations.
  10.  Put the dates below in a calendar, and review quarterly.

The calendar: dates that matter next

Date What happens Where
November 16, 2026 California agencies due to deliver recommendations to the Governor on amending state AI safety laws (Executive Order N-9-26) California
December 2, 2026 Two new EU bans (AI-generated non-consensual intimate imagery and child sexual abuse material) apply; machine-readable marking duty applies to generative systems already on the market EU
January 1, 2027 Colorado’s rewritten automated-decision law takes effect, per two trackers (check, since sources disagree); New York and Illinois frontier-model laws queued for 2027 U.S. states
About January 22, 2027 or later Earliest end of South Korea’s indicated one-year grace period before fines South Korea
August 2, 2027 National AI regulatory sandboxes due to be operational EU
December 2, 2027 High-risk obligations for stand-alone (Annex III) systems EU
August 2, 2028 High-risk obligations for AI embedded in regulated products (Annex I) EU
August 2, 2030 Legacy high-risk systems used by public authorities EU

Myths and facts

Claim you may see What the evidence says
“The EU delayed the AI Act, so nothing applies until 2027.” Only the high-risk rules moved. Bans, AI literacy, GPAI duties, transparency and Commission enforcement are already in effect, and more dates land on December 2, 2026.
“The U.S. banned state AI laws.” No. Executive orders and a litigation task force target state laws, but no federal preemption statute exists and state laws remain in force.
“The U.S. has no AI rules.” There is no federal AI statute, but state laws in California, Texas, Illinois, New York City and elsewhere are in force.
“China has no AI law, so there are no rules.” China has binding targeted rules, including mandatory labeling since September 2025.
“South Korea’s law is not enforced.” It is in force since January 22, 2026, but fines are paused for at least a year.
“Japan’s AI law is strict.” It is a promotion law with no penalties.
“If I only use an AI API, none of this applies to me.” The provider carries the GPAI model duties, but you may still have transparency, data-protection and, depending on use, high-risk obligations.
“Labeling AI content is optional.” In the EU, China, South Korea and India, labeling or disclosure duties exist in some form.

Frequently asked questions

Did the EU delay the AI Act?

It delayed the high-risk rules: stand-alone high-risk systems to December 2, 2027 and AI embedded in regulated products to August 2, 2028, under Regulation (EU) 2026/1744, which entered into force on July 27, 2026. Other obligations, including transparency duties and Commission enforcement of the GPAI rules, have applied since August 2, 2026.

What happens on December 2, 2026 in the EU?

Two new bans on AI-generated non-consensual intimate imagery and child sexual abuse material apply, and generative AI systems already on the market must meet the machine-readable marking duty, according to several trackers.

Does the EU AI Act apply to non-EU companies?

Summaries say it applies to providers and deployers outside the EU where their AI is placed on the EU market or its output is used in the EU. Check your own situation with a lawyer.

Is there a federal AI law in the United States?

Not as of early November 2026, according to a legal tracker. The federal government has issued executive orders and a policy framework, and state laws are in force.

Can the federal government override state AI laws?

Not by executive order alone. Executive Order 14365 created a litigation task force and sought legislation, but the trackers we read found no preemption statute.

Which U.S. state AI laws are in force?

California’s SB 53 and other laws, Texas’s TRAIGA, Illinois’s HB 3773, New York City’s Local Law 144 and others, as reported by trackers. Colorado’s status is disputed between sources.

What is South Korea’s AI Basic Act?

A comprehensive AI framework law in force since January 22, 2026, with duties for high-impact AI and generative AI labeling, a domestic-representative requirement for larger foreign providers and fines up to KRW 30 million, paused for at least a year.

Does China require labels on AI-generated content?

Yes. Since September 1, 2025, providers must add visible labels where content could mislead and embed machine-readable metadata, and platforms have verification duties, according to Latham & Watkins’ summary.

Do I need to label AI-generated images on my blog?

In the EU, certain AI-generated or manipulated content such as deepfakes must be labeled, and other jurisdictions have similar rules. Even where not strictly required, clear labeling and an AI-use disclosure are good practice.

What is the biggest mistake companies make?

According to commentary we read, pausing compliance work because of the EU delay. Several obligations are already in force, and more apply on December 2, 2026.

The 2026 story of AI regulation is not a single global trend but three different dynamics. In Europe, the law is real and enforceable, with its toughest high-risk rules delayed and the rest already live. In the United States, there is no federal statute, and states and the federal government are in a tug-of-war. In Asia, the approaches diverge: China regulates content and services through targeted rules, South Korea has a comprehensive law with a gentle start, and Japan, India and Singapore lean on guidance.

For anyone building, publishing or deploying AI, the practical response is to stop asking whether “AI regulation” applies and ask which rules apply to your role, your users and your country. Start with transparency and labeling, which almost every system requires, inventory your AI uses, and keep the dates in your calendar, starting with December 2, 2026. Verify everything against the official sources and take legal advice, because this is general information, not legal advice.

Sources

URLs below were returned during our November 15, 2026 research. Almost all are law-firm, consultancy or tracker sites; verify dates and fines against the official texts (such as EUR-Lex, Korea’s law database and China’s Cyberspace Administration) before relying on them.

1. The Digital AI Omnibus: Proposed deferral of high-risk AI obligations. DLA Piper (Omnibus background, 2 Dec 2027 proposal). 

2. EU AI Act Omnibus Agreement: Postponed High-Risk Deadlines and Other Key Changes. Gibson Dunn (deferral is not a dismantling; Article 50 stays). 

3. The EU AI Act After the AI Omnibus: The Timeline That Now Applies. McKenna Consultants (timeline table; Commission statement on entry into force). 

4. EU AI Act Deadlines After the Digital Omnibus. Technology.org, October 13, 2026 (December 2026 deadlines; sandboxes; legacy systems). 

5. EU AI Act Digital Omnibus: High-Risk Deadline Now Dec 2027. AI Policy Desk (Parliament vote, Council adoption and publication dates). 

6. EU AI Act’s High-Risk Deadline: Deferred, Not Cancelled. Cloud Security Alliance research note (Regulation 2026/1744 dates). 

7. The EU AI Act: What Actually Applies From August 2026. Digital Applied (obligation table). 

8. EU AI Act Digital Omnibus 2026: What It Changes. regulation-ai.eu. 

9. AI Laws in the US: State-by-State Tracker. RecordingLaw (no federal statute as of November 4, 2026; Task Force date; California EO N-9-26). 

10. US AI Regulation Outlook: Board Briefing. NHIMG (EO 14365, status at November 11, 2026). 

11. United States AI Policy 50-State Tracker. George Mason University Center for Assurance Research and Engineering (state status as of October 17, 2026). 

12. Trump AI Executive Order and State Law Preemption. AI Laws by State (Colorado litigation; preemption risk). 

13. US State AI Laws: What Enterprise Deployers Owe Now. Deepak Gupta (laws in force as of October 30, 2026). 

14. AI Laws of the World: United States. DLA Piper (Task Force; TRAIGA; SB 53).

15. AI Regulation in APAC: Diverging Approaches Across the Region. Latham & Watkins (South Korea, China labeling). 

16. AI Regulation in 2026: What Applies Now. iTechGuides (Japan, India, Singapore, EU December 2). 

17. How 8 Countries Regulate AI in 2026. AskAjay (China Cybersecurity Law amendment; Korea fines). 

18. AI Regulations by Country. ComplyJet (country table; Vietnam). 

19. Global AI Regulation in 2026: EU AI Act vs USA, China, South Korea, Japan and the UK. Blocksize (UK; Japan’s basic plan). 

20. AI Regulation 2026: EU Enforces, US Waits. FutureCareerInfo (three models). 

21. AI Compliance: How Rules Differ by Country. TechBloat (Korea compute threshold; India; Singapore). 

22. AI Policy in Asia: National Strategies and Regulatory Frameworks. Digital in Asia (Japan basic plan July 2026, Korea action plan, Taiwan). 

23. Asia AI Regulation Tracker. AsiaAI.FYI (Japan, Korea, Taiwan, Vietnam, Singapore).Â