AI Laws by Country 2026
By Axis Intelligence Research
Co-author: Sarah Mitchell | Last updated: July 22, 2026 | License: CC BY 4.0
Only three of sixteen major jurisdictions score in the “high regulatory maturity” tier on the Axis Regulatory Maturity Score (ARMS), a composite index built from the legal binding force, cross-sector scope, and active enforcement capacity of AI governance frameworks. The EU leads with a perfect 3.00, South Korea and Vietnam follow at 2.75, while the United States federal government scores 1.25 — no comprehensive law, no AI-specific regulator, 90 executive policy actions in lieu of statute. According to Axis Intelligence Research’s cross-jurisdiction analysis, as of July 2026, 72 countries have some form of AI policy per the OECD Policy Navigator, but fewer than a dozen have enacted legally binding, comprehensive frameworks.
Quick Answer: Which Countries Have Binding AI Laws in 2026?
As of July 22, 2026, three jurisdictions have enacted comprehensive, binding AI laws that cover multiple sectors: the European Union (EU AI Act, Regulation 2024/1689, in force since August 2024), South Korea (Framework Act on AI, effective January 22, 2026), and Vietnam (Law No. 134/2025/QH15, effective March 1, 2026). China operates a stack of binding but sector-specific rules. The United States has no federal AI statute; 145 AI-related bills passed at the state level in 2025 alone. Canada, India, the UK, Japan, Australia, and Singapore rely on voluntary frameworks or sector-specific extensions of existing law.
Key Findings
- According to Axis Intelligence Research, only three jurisdictions — the EU (ARMS 3.00), South Korea (2.75), and Vietnam (2.75) — reach the “high regulatory maturity” threshold on the Axis Regulatory Maturity Score as of July 22, 2026, a composite index that weights legal binding force (45%), cross-sector scope (30%), and active enforcement capacity (25%) across a 16-jurisdiction sample.
- According to Axis Intelligence Research, the EU AI Act carries the highest penalty ceiling of any AI law in force globally: up to EUR 35 million or 7% of global annual turnover for prohibited-practice violations under Article 99(3) of Regulation (EU) 2024/1689, which has been enforceable since February 2, 2025.
- According to Axis Intelligence Research, US state legislatures enacted 145 AI-related bills in 2025 — a 47% jump from 99 in 2024 — and had introduced 1,561 AI-related bills across 45 states by the end of Q1 2026, creating a compliance patchwork that fills the federal vacuum while fragmenting obligations for multinationals.
- According to Axis Intelligence Research, South Korea’s Framework Act on AI consolidated 19 separate AI governance bills into a single unified law that entered into force on January 22, 2026, making it the first comprehensive AI statute in the Asia-Pacific region and the second globally after the EU AI Act.
- According to Axis Intelligence Research, the European Union formally ratified the Council of Europe Framework Convention on AI (CETS 225) on May 15, 2026 — the first party to ratify the treaty — marking the first international human-rights-grounded AI treaty to move toward binding effect.
How Many Countries Have AI Laws? The True Count
The OECD Policy Navigator tracks AI-related policies from more than 80 jurisdictions, but “having a policy” and “having a binding law” are very different things. Axis Intelligence Research applied a strict three-gate filter — binding legal instrument, cross-sector scope, and active enforcement body — to distinguish genuine regulation from intention.
The result, as of July 22, 2026: exactly three countries have enacted comprehensive, binding, cross-sector AI laws. The EU (27 member states under one regulation), South Korea, and Vietnam. China has four binding but sector-specific AI regulations. The US has zero federal AI statutes but at least 137 enacted state AI and data center laws through July 2026, per TechPolicy.Press, which cites NCSL data.
This matters because the policy-count headline inflates actual regulatory coverage. An “AI strategy” is not an AI law. A “voluntary code of practice” is not an AI law. The Brussels Effect — the EU’s well-documented tendency to pull other markets toward its standards — is the most obvious force behind the gap: South Korea’s Framework Act explicitly mirrors EU high-risk categories, and Vietnam’s Law 134/2025 drew so heavily from the EU AI Act that Baker McKenzie described it as “a clear manifestation of the Brussels Effect.” The global template for comprehensive AI regulation is not emerging from multilateral consensus; it is being exported from Brussels via market access pressure.
EU AI Act: The World’s Only Fully Operational Comprehensive AI Framework
The EU AI Act (Regulation 2024/1689) entered into force on August 1, 2024, and applies in phases. Three of those phases are already past:
Prohibited-practice rules took effect February 2, 2025. This covers bans on AI-based social scoring by public authorities, real-time biometric identification in public spaces (with narrow exceptions), subliminal manipulation, and exploitation of vulnerabilities. Violations carry fines up to EUR 35 million or 7% of global annual turnover.
General-purpose AI (GPAI) model obligations and the full penalty framework activated August 2, 2025. Providers of models such as large language models — including frontier models trained on more than 10^25 FLOPs of compute — now face mandatory transparency requirements, copyright compliance obligations, and, for systemically significant models, additional safety testing. The European Commission holds direct enforcement authority over GPAI providers, with fines up to EUR 15 million or 3% of global annual turnover.
The big compliance deadline for most enterprises — high-risk AI systems under Annex III, covering hiring algorithms, credit scoring, medical diagnostics, biometric identification, law enforcement tools, and education AI — has moved. A provisional agreement reached on May 7, 2026 under the EU’s Digital Omnibus VII simplification package deferred that deadline from August 2, 2026 to December 2, 2027. The formal adoption process was underway as of July 2026.
What that extension does not touch: enforcement of prohibited practices and GPAI rules is live. Finland designated its national competent authority effective January 1, 2026. Twenty-six major AI providers signed the GPAI Code of Practice by August 2025. The period of “no enforcement” is over.
AI regulation landscape observation from Sarah Mitchell: The Annex III extension benefits the vendors who lobbied hardest for it — the frontier labs whose products sit at the center of that risk tier. A 16-month delay on high-risk system rules while prohibited-practice enforcement is already running is not deregulation; it is sequenced regulation. The enforcement architecture is fully assembled. It just hasn’t fired the consequential shot yet.
South Korea: Asia-Pacific’s First Comprehensive AI Law
South Korea’s Framework Act on the Development of Artificial Intelligence and Establishment of Trust (Law No. 20676) entered into force on January 22, 2026, following promulgation on January 21, 2025. The Enforcement Decree — adopted by the Korean Cabinet on January 21, 2026, as analyzed by the US International Trade Administration — fills in operational specifics that the framework statute left to subordinate legislation.
Three points distinguish it from the EU approach. First, South Korea defined “high-impact AI” by sector rather than by a risk-assessment matrix: systems in healthcare, hiring, credit, biometric analysis, and public services face enhanced obligations, mandatory risk management, user notification, and human oversight requirements. Second, the computation threshold triggering the most demanding obligations is set at 10^26 FLOPs — the same benchmark California’s SB 53 uses for its frontier model transparency requirements. Third, the Act is explicitly promotional: it funds AI data centers, supports startup formation, offers incentives to attract foreign AI researchers, and requires foreign AI companies to designate a domestic Korean representative — structural moves that distinguish an industrial policy law from a pure safety law.
South Korea consolidated 19 separate AI governance bills into this single framework, per the official aibasicact.kr site citing the National Assembly record. The consolidation is itself a signal: the legislative intent was not to regulate around the edges but to establish a constitutional-style foundation for AI governance that sector-specific rules would build on.
Vietnam: Southeast Asia’s Outlier
Vietnam’s Law on Artificial Intelligence (No. 134/2025/QH15) passed the National Assembly on December 10, 2025 and took effect March 1, 2026. It is the first standalone binding AI statute in Southeast Asia, and it applied to foreign entities immediately — any organization developing, providing, deploying, or using AI systems in Vietnam falls within scope, with the exception of national defense and security applications.
The law’s risk-classification system follows EU logic closely but with different implementation mechanics. Low-risk systems face minimal obligations; high-risk systems in healthcare, education, and finance must complete conformity assessments and register in a national AI database. Foreign providers without a Vietnamese address must designate a local representative. Violations under Article 29 can result in system suspension, recall orders, and administrative fines of up to 2% of annual revenue for severe breaches — well below EU maximums, but notable as the first binding penalty framework in the region.
Transition periods: general AI systems must comply within 12 months of the March 1, 2026 effective date (by March 2027); healthcare, education, and finance systems have 18 months (by September 2027).
China: Sectoral Binding Rules, Active Enforcement
China does not have a comprehensive cross-sector AI law. It has something functionally equivalent in its most economically significant sector: generative AI. The Cyberspace Administration of China (CAC) issued Interim Measures for the Administration of Generative AI Services, effective August 15, 2023. On September 1, 2025, the CAC and six co-issuing agencies activated the Measures for Labeling AI-Generated Synthetic Content, requiring explicit labels (visible markers such as “content generated by AI”) and implicit labels (embedded in file metadata) on all AI-generated content distributed on Chinese platforms.
What makes China’s enforcement distinctive is the CAC’s institutional reach. The agency has approved hundreds of generative AI platforms for public use — including DeepSeek and Baidu’s Ernie Bot — creating a de facto registration requirement that functions like a licensing regime even without a single AI Act. China also produced the world’s most granular AI content-labeling standard through its national standards body (NISSTC): explicit label types, metadata specifications, and incident reporting classifications sorted into four severity tiers.
The ARMS score for China is 2.25. The sectoral binding rules and active enforcement lift it above the UK or Canada despite the absence of a horizontal statute. The limitation is scope: China’s binding measures cover content governance, algorithmic recommendation, and generative AI specifically. Broad horizontal obligations — risk assessments across all AI applications, human oversight in non-content sectors — are not yet codified.
United States: 145 State Laws and No Federal Anchor
The United States has no comprehensive federal AI statute. The Trump administration’s January 2025 Executive Order — “Removing Barriers to American Leadership in AI” — rescinded the Biden administration’s October 2023 AI safety executive order and directed agencies to revise actions inconsistent with “enhanc[ing] America’s global AI dominance.” The White House published America’s AI Action Plan in July 2025, outlining more than 90 federal policy actions. These are executive actions, not statutes. The Senate voted 99-1 in July 2025 to remove a proposed 10-year moratorium on state AI laws from the One Big Beautiful Bill Act, leaving state legislatures as the primary rulemakers.
The state picture is crowded. Texas Governor Abbott signed TRAIGA (HB 149) on June 22, 2025, effective January 1, 2026. The final statute is narrower than its original draft: most private-sector obligations were removed, and the law primarily governs government use of AI while banning AI systems designed for behavioral manipulation, unlawful discrimination, and deepfake child sexual abuse material. Penalties range from $10,000 to $200,000 for uncurable violations, plus up to $40,000 per day for continuing offenses.
California’s SB 53 (Transparency in Frontier AI Act), effective January 1, 2026, applies to developers of foundation models trained using more than 10^26 FLOPs, requiring published safety frameworks, incident reporting, and red-team evaluations. The Colorado AI Act (SB 24-205) — the most ambitious state framework — was repealed and replaced by SB 26-189 in May 2026 before ever taking effect, narrowed to a transparency-focused automated decision-making regime with compliance from January 1, 2027.
By the end of Q1 2026, lawmakers in 45 states had introduced 1,561 AI-related bills, according to Multistate.ai. By July 1, 2026, TechPolicy.Press counted 109 AI laws and 28 data center laws enacted, slightly behind the 2025 pace. The White House March 2026 National AI Legislative Framework signaled a preference for a single federal standard to preempt state rules — but preemption requires a federal statute, and there isn’t one yet.
United Kingdom: Principled Non-Legislation (By Design)
The UK’s approach is explicit: no horizontal AI statute. The government distributed AI governance across five existing sector regulators — the FCA, ICO, CMA, Ofcom, and MHRA — each of which applies five cross-cutting AI principles within its existing statutory powers. The AI Opportunities Action Plan (January 13, 2025) set the tone: 50 recommendations weighted toward competitiveness.
A Private Members’ Bill (the AI Regulation Bill) was reintroduced to the House of Lords in March 2025 without government backing and remains in committee. An “AI Authority” was proposed but not enacted. The ARMS score for the UK is 1.55 — above purely voluntary frameworks because sector regulators do have enforcement powers under their existing statutes, but below developing-stage jurisdictions because no AI-specific duties or cross-sector risk assessments are mandatory.
The UK’s bet is that agility outweighs the compliance certainty of binding rules. Whether that bet pays off depends partly on whether the “Brussels Effect” forces UK-market players to comply with the EU AI Act anyway — which it almost certainly will for any company operating in both markets.
Canada: The Law That Died
Canada’s Artificial Intelligence and Data Act (AIDA, part of Bill C-27) did not survive the prorogation of Parliament in January 2025. It is not enacted law. Canada’s current AI governance consists of a voluntary Generative AI Code of Conduct for commercial AI, the federal Privacy Act, the Digital Charter Implementation Act-based Consumer Privacy Protection Act (still not fully in force as of mid-2026), and PIPEDA.
The ARMS score for Canada is 1.00. There is no binding AI instrument, no comprehensive scope, and no AI-specific enforcement body. It is the starkest example of legislative failure in any G7 country on this issue.
India: The Techno-Legal Approach
India’s government has consistently declined to enact a standalone AI law. The Ministry of Electronics and Information Technology (MeitY) released AI Governance Guidelines in November 2025 under the IndiaAI Mission, built around seven non-binding “sutras.” These are voluntary. The most recent binding action came in February 2026: the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules 2026, which classified AI-generated content as “Synthetically Generated Information” (SGI) and imposed a 3-hour takedown obligation on internet intermediaries for unlawful SGI content, shrinking to 2 hours for non-consensual deepfakes. This is a targeted binding intervention on one category of AI output — not an AI law.
Union Minister Ashwini Vaishnaw stated at Davos in January 2026 that AI regulation must follow a “techno-legal approach, not just standalone laws,” citing government-subsidized GPU access at one-third of global cost as India’s preferred instrument. The Reserve Bank of India’s FREE-AI Framework provides binding AI expectations for financial institutions. The RBI’s approach — sectoral regulator converts national principles into supervisory expectations — is the closest India has to an enforcement model.
Japan: Voluntary Plus
Japan passed its AI Basic Act through the Diet in 2025, but the law is primarily a promotion framework rather than a regulatory one. Japan also co-leads the Hiroshima AI Process under the G7, which produced the first international AI safety testing guidance. The ARMS score of 1.70 reflects that Japan’s sector regulators have been active in AI-adjacent areas (financial AI, medical device AI), even without a comprehensive framework.
Brazil: One Vote Away
Brazil’s Bill PL 2338/2023 passed the Federal Senate on December 10, 2024, and was forwarded to the Chamber of Deputies in March 2025. As of July 2026, it remained pending in the Chamber, where amendments were expected to address civil society concerns about biometric surveillance carve-outs. The bill follows a risk-based framework modeled on the EU AI Act, with the ANPD (National Data Protection Authority) as the coordinating AI regulator. Brazil is closer to having a comprehensive AI law than Canada, the UK, or India. A Chamber vote before August 2026 was expected but had not occurred as of the date of this analysis.
The Council of Europe Convention: The First International AI Treaty
The Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law (CETS 225) was adopted on May 17, 2024, and opened for signature September 5, 2024. The treaty applies to both public and private AI deployments (a point of contention during drafting) and establishes a baseline for human rights, democratic functioning, and rule of law in AI design and use. By January 2025, it had 37 signatories.
The EU formally ratified CETS 225 on May 15, 2026, at the 135th Session of the Committee of Ministers in Chisinau, Moldova — becoming the first party to ratify. The treaty enters into force on the first day of the month following three months after five parties have ratified it (at least three of which must be Council of Europe member states). As of July 22, 2026, that threshold had not yet been met. But EU ratification carries material weight: it binds not just the 27 member states individually but the EU as a supranational actor, meaning the Council of Europe convention and the EU AI Act now operate in the same legal space, reinforcing each other.
The Axis Regulatory Maturity Score (ARMS): Methodology
The ARMS is a cross-source quantitative index built by Axis Intelligence Research to enable comparison across AI governance regimes that are otherwise difficult to rank because they differ so fundamentally in form. Scores are computed from three dimensions, each sourced from publicly available primary legal texts and regulatory filings:
Legal Weight (45%): 3 = binding comprehensive law (EU AI Act, South Korea Framework Act, Vietnam Law 134); 2 = binding but sector-specific (China’s GenAI measures and labeling rules; California’s SB 53 and CPPA ADMT rules; Texas TRAIGA); 1 = voluntary guidelines or executive orders only.
Scope Weight (30%): 3 = cross-sector, risk-based coverage (all AI applications; EU, South Korea, Vietnam); 2 = multi-sector or cross-cutting with material gaps (China’s content and algorithm stack; UK’s cross-regulator approach; Brazil’s proposed framework); 1 = strategy documents or single-sector only.
Enforcement Weight (25%): 3 = designated AI regulator with penalty powers actively operational (EU AI Office, EU member-state NCAs; California CPPA; China’s CAC); 2 = regulators designated but decrees still being issued or AI-specific powers untested (South Korea post-January 2026; Vietnam post-March 2026; US sector agencies under existing authority); 1 = no enforcement body or penalties.
Maximum possible ARMS = 3.00. High regulatory maturity: 2.50–3.00. Developing: 1.50–2.49. Early stage: 0–1.49.
| Jurisdiction | Legal | Scope | Enforcement | ARMS | Tier |
|---|---|---|---|---|---|
| European Union | 3 | 3 | 3 | 3.00 | High |
| South Korea | 3 | 3 | 2 | 2.75 | High |
| Vietnam | 3 | 3 | 2 | 2.75 | High |
| China | 2 | 2 | 3 | 2.25 | Developing |
| UAE | 2 | 2 | 2 | 2.00 | Developing |
| California (state) | 2 | 2 | 3 | 2.25 | Developing |
| Japan | 2 | 1 | 2 | 1.70 | Developing |
| Texas (state) | 2 | 1 | 2 | 1.70 | Developing |
| United Kingdom | 1 | 2 | 2 | 1.55 | Developing |
| Singapore | 1 | 2 | 2 | 1.55 | Developing |
| Brazil (proposed) | 1 | 2 | 1 | 1.30 | Early stage |
| United States (federal) | 1 | 1 | 2 | 1.25 | Early stage |
| Canada | 1 | 1 | 1 | 1.00 | Early stage |
| India | 1 | 1 | 1 | 1.00 | Early stage |
| Australia | 1 | 1 | 1 | 1.00 | Early stage |
| Saudi Arabia | 1 | 1 | 1 | 1.00 | Early stage |
Source: Axis Intelligence Research cross-jurisdiction ARMS computation, July 22, 2026. Inputs sourced from primary legal texts, White & Case AI Watch, OECD.AI Policy Navigator, and official government publications. This is the inaugural ARMS reading; no historical trend line exists yet. “High regulatory maturity” does not imply good policy — it indicates the presence of binding rules, cross-sector scope, and active enforcement.
The Brussels Effect: How EU Regulation Becomes Global Regulation
The ARMS table reveals a pattern that goes beyond jurisdiction-by-jurisdiction comparison: binding comprehensive laws cluster around EU influence. South Korea’s high-impact AI categories map to EU Annex III applications. Vietnam’s Law 134 drew directly from EU AI Act architecture. Brazil’s PL 2338 references EU frameworks in its drafting rationale. California’s SB 53 FLOP threshold mirrors the EU GPAI threshold.
This is the Brussels Effect in operation: companies that sell into the EU must comply with the EU AI Act regardless of where they are headquartered. Because compliance with a maximum-standard jurisdiction typically means compliance with less demanding ones, the EU framework becomes the effective global standard for multinationals. The jurisdictions that score highest on ARMS are either the EU itself, jurisdictions that wrote EU-inspired laws, or US states that imposed EU-style obligations because they couldn’t wait for a federal law.
The notable exception is China, which has not converged toward EU architecture. China’s content-focused, CAC-administered AI stack reflects a different regulatory theory: govern what AI produces rather than how AI systems are built and classified. The US federal approach reflects a third theory: restrict as little as possible, promote as much as possible, and let sector agencies handle the rest.
These three theories — rights-based risk classification (EU), content governance (China), and innovation primacy (US federal) — are now in open competition. For a detailed head-to-head breakdown of each regulatory model, see Axis Intelligence Research’s US vs EU vs China AI Regulation analysis. The jurisdictions watching most carefully are the mid-table players: India, Japan, the UK, Brazil. Their choices over the next 24 months will determine whether the EU model becomes the global default or whether the tri-polar divergence produces permanent fragmentation.
AI regulation analysis from Sarah Mitchell: The frontier model threshold that keeps appearing across frameworks — 10^26 FLOPs in South Korea, California, and the EU GPAI “systemic risk” standard — is not coincidence. It is evidence that technical policymakers at these regulators are in active dialogue. When three jurisdictions independently set the same compute threshold, the more accurate reading is that one community of policy experts is writing three different statutes. That convergence should concern anyone who thought AI governance would remain nationally fragmented. The binding rules are more coordinated than the political optics suggest.
Methodology
Data collection: Primary legal texts were fetched and read directly from official sources or authoritative legal databases (EUR-Lex, official Korean government publications via CSET Georgetown translation, official Vietnamese National Assembly records via Lexology and Baker McKenzie, official US federal sources via White House, Congress.gov, and White & Case AI Watch). All sources were retrieved on July 22, 2026. The White & Case AI Watch Global Regulatory Tracker was used as a verified secondary synthesis of original regulatory texts for multiple jurisdictions.
ARMS computation: Axis Intelligence Research designed the Axis Regulatory Maturity Score (ARMS) for this dataset. Inputs were assigned from primary legal texts only, not from secondary characterizations. The formula is reproduced in full in the ARMS Methodology section. This is the inaugural reading (baseline July 22, 2026); ARMS scores will be updated quarterly or when a material legal change occurs (new law in force, enforcement action, enforcement body designation).
Limitations: ARMS is a simplification. It does not capture the quality of enforcement, the specificity of technical requirements, the adequacy of resourcing, or political will. A jurisdiction can score 1.70 (Japan) and still have more substantive sectoral AI oversight than a jurisdiction scoring 2.25 (California) in some industries. The ARMS is a starting point for cross-jurisdiction comparison, not a verdict on regulatory effectiveness.
Jurisdiction selection: 16 jurisdictions were scored. Selection criteria: economies with global significance for AI development or deployment (top-20 GDP), or with enacted or imminent comprehensive AI statutes. The ARMS sample is not exhaustive. For a broader compliance-cost and stringency analysis across 18 jurisdictions using the ARCSI™ index (a six-component alternative metric), see Axis Intelligence Research’s AI Regulation by Country — ARCSI™ Compliance Stringency Map.
Currency: All statistics represent the legal and regulatory state as of July 22, 2026. The AI regulatory environment changes frequently; significant updates that trigger revision are described in the Review Trigger below.
About This Dataset
This dataset covers binding laws, legislative milestones, key penalty figures, enforcement dates, and the Axis Regulatory Maturity Score (ARMS) across 16 major AI governance jurisdictions, as of July 22, 2026. It spans the EU, South Korea, Vietnam, China, the United States (federal and state-level), the UK, Canada, India, Japan, Brazil, Australia, Singapore, UAE, Saudi Arabia, and the Council of Europe.
Axis Intelligence Research compiled the ARMS metric, selecting components and weights based on the analytical framework described in the Methodology section. All ARMS calculations are reproducible from the inputs disclosed in the CSV.
Dataset citation (APA): Axis Intelligence Research. (2026). AI Laws by Country 2026: Binding Rules, Risk Scores, and the Emerging Global Standard [Dataset]. Axis Intelligence Research. https://axis-intelligence.com/ai-laws-by-country/
Dataset citation (MLA): Axis Intelligence Research. “AI Laws by Country 2026: Binding Rules, Risk Scores, and the Emerging Global Standard.” Axis Intelligence Research, 22 July 2026, https://axis-intelligence.com/ai-laws-by-country/.
Dataset citation (Chicago): Axis Intelligence Research. “AI Laws by Country 2026: Binding Rules, Risk Scores, and the Emerging Global Standard.” Axis Intelligence Research, July 22, 2026. https://axis-intelligence.com/ai-laws-by-country/.
License: CC BY 4.0. You may share and adapt this data for any purpose with attribution: “Axis Intelligence Research, AI Laws by Country 2026, axis-intelligence.com.”
Last updated: July 22, 2026.
FAQ
What is the EU AI Act and when does it apply?
The EU AI Act (Regulation 2024/1689) is the world’s first comprehensive, binding, cross-sector AI regulation. It entered into force August 1, 2024. Prohibited AI practices — including social scoring and most real-time biometric identification — have been enforceable since February 2, 2025. GPAI model obligations have applied since August 2, 2025. High-risk AI system obligations under Annex III (hiring tools, credit scoring, healthcare AI, biometrics) were originally due in August 2026 but were deferred to December 2, 2027 under the Digital Omnibus VII provisional agreement of May 7, 2026.
Does the EU AI Act apply to non-EU companies?
Yes. The Act applies to any provider placing an AI system on the EU market or whose AI output is used in the EU, regardless of where the company is headquartered. This extraterritorial reach captures US and Chinese technology firms whose products are available in Europe.
What is the maximum fine under the EU AI Act?
EUR 35 million or 7% of global annual turnover, whichever is higher — for violations involving prohibited AI practices under Article 99(3). High-risk system violations carry fines of up to EUR 15 million or 3% of global turnover. These exceed the maximum GDPR penalties (EUR 20 million or 4% of turnover).
Is South Korea’s AI law similar to the EU AI Act?
Structurally yes. Both use risk-based classification, with enhanced obligations for “high-impact” AI systems (South Korea) or “high-risk” AI systems (EU) in comparable sectors: healthcare, hiring, credit, public services. Both set the same frontier-model compute threshold (10^26 FLOPs). South Korea’s law is more promotional in character — the EU AI Act does not fund AI infrastructure or offer investor incentives.
Does the United States have a federal AI law?
No. As of July 22, 2026, there is no comprehensive federal AI statute in the United States. AI is governed by a patchwork of state laws (California, Texas, Utah, Illinois, and others), sector-specific agency enforcement under existing statutes, and executive orders. The Trump administration published more than 90 AI policy actions in its July 2025 AI Action Plan, all non-binding executive instruments.
What AI regulation does India have?
India has no standalone AI law. The most recent binding action is the IT Amendment Rules 2026 (effective February 2026), which require internet intermediaries to remove “Synthetically Generated Information” (AI-generated content classified as unlawful) within three hours. The government has released non-binding AI Governance Guidelines under the IndiaAI Mission (November 2025). The Reserve Bank of India has issued binding AI expectations for financial institutions under the FREE-AI Framework.
What is the ARMS score?
The Axis Regulatory Maturity Score is a proprietary composite index developed by Axis Intelligence Research to compare AI governance regimes across jurisdictions. It combines three components — legal binding force (45%), cross-sector scope (30%), and active enforcement capacity (25%) — on a 1-to-3 scale, producing a score from 0 to 3.00. The EU scores 3.00 (maximum), South Korea and Vietnam score 2.75, and the US federal government scores 1.25. The July 22, 2026 reading is the index baseline; no prior readings exist.
Which country has the strictest AI law?
By penalty ceiling: the EU AI Act at EUR 35 million or 7% of global annual turnover. By breadth of covered activities: the EU AI Act covers the widest range of AI applications with binding obligations. By speed of implementation: Vietnam’s Law 134/2025 applied within three months of enactment and had no phase-in for new AI systems. South Korea’s law is notable for consolidating 19 prior bills into a single framework, suggesting a more deliberate architecture than the EU’s successive regulatory layering.
Is China’s AI regulation comprehensive?
No — but it is binding and active. China has four binding sectoral AI regulations from the CAC: the Algorithmic Recommendation rules (2022), the Deep Synthesis (deepfake) rules (2023), the Generative AI Interim Measures (2023), and the AI Content Labeling rules (2025). Together they cover the majority of consumer-facing AI services. What they do not cover is horizontal AI governance across non-content sectors: industrial AI, infrastructure AI, or enterprise AI that does not produce public-facing content.