US vs EU vs China AI Regulation 2026
By Axis Intelligence Research
Co-author: Sarah Mitchell | Last updated: July 19, 2026 | License: CC BY 4.0
The EU’s AI Act carries fines of up to €35 million or 7% of global annual turnover — the steepest penalty ceiling in the history of technology regulation. China has already enforced its generative AI filing mandate against 868 registered services as of April 2026. The United States, operating without a single federal AI statute, has watched 145 state laws pass in 2025 alone. Three regulatory architectures. Three incompatible compliance burdens. One set of globally deployed AI systems caught between them.
Quick Answer
The EU operates the world’s most prescriptive AI regulatory framework: Regulation (EU) 2024/1689 creates a four-tier risk hierarchy, bans 14 AI practices outright, and carries penalties up to 7% of global turnover (Article 99). China runs a layered enforcement system anchored in its 2023 Generative AI Measures, with 868 registered services and swift administrative action — but no unified statute yet. The United States has no federal AI law; Executive Order 14365 (December 2025) and a June 2026 follow-up establish voluntary frameworks and seek to preempt state regulation, while 45 states introduced 1,561 AI-related bills in the first half of 2026 alone.
Key Findings
- According to Axis Intelligence Research’s Regulatory Divergence Index™ (RDI™), the EU scores 88.0 out of 100 on regulatory comprehensiveness, China scores 74.0, and the United States scores 32.4 — a 55.6-point gap between the world’s most and least prescriptive major AI regimes as of July 2026.
- The EU AI Act’s maximum fine of €35 million or 7% of global annual turnover (Article 99) exceeds GDPR’s ceiling, making it the highest percentage-based penalty regime in EU digital regulation, per the Act’s own text as confirmed by national implementing legislation.
- China’s Cyberspace Administration of China recorded 868 cumulative generative AI service registrations as of April 30, 2026, with 530 additional application filings — the only publicly accessible national registry of its kind globally (Wedoany / CAC announcement, May 13, 2026).
- In Pew Research Center’s 2025 survey of 25 countries, a median of 53% of adults trusted the EU to regulate AI — compared to 37% for the United States and 27% for China, the largest trust differential recorded between major AI powers (Pew Research Center, October 2025).
- US states introduced 1,561 AI-related bills in the first 45-state legislative survey through March 2026, already surpassing the entire 2024 total, while the only federal AI-specific statute enacted in 2025 was the Take It Down Act covering nonconsensual deepfakes (MultiState, 2026).
What Framework Governs AI in Each Jurisdiction?
The EU: One Binding Regulation, Four Risk Tiers
The EU AI Act (Regulation (EU) 2024/1689) entered into force on August 1, 2024. It is the world’s first comprehensive binding AI law. It applies to any developer, deployer, or importer of AI systems placed on the EU market or used within the EU — regardless of where the company is headquartered.
The framework organizes AI applications into four risk tiers:
- Unacceptable risk (prohibited): Social scoring by public authorities, real-time biometric identification in public spaces, subliminal manipulation, emotion recognition in workplaces and schools. These bans have been in force since February 2, 2025.
- High risk: AI in hiring, credit, healthcare, law enforcement, critical infrastructure, and education. Full conformity assessment and EU database registration requirements.
- Limited risk: Chatbots, deepfakes, and emotion-recognition tools subject to transparency disclosure obligations (enforceable since August 2, 2025 under Article 50).
- Minimal risk: Most commercial AI — no mandatory obligations beyond general good practice.
The European AI Office, established within the European Commission, directly enforces obligations for general-purpose AI (GPAI) model providers. National market surveillance authorities enforce the rest.
Sarah Mitchell notes: the EU’s four-tier architecture was designed in 2021 when “large language model” was barely a category. Reality has been less tidy. Studies from the appliedAI Enterprise Survey and the German AI Association consistently find that 18–50% of deployed enterprise AI systems qualify as high-risk — three to ten times the Commission’s original 5–15% estimate. That undercount is not a minor calibration error; it is the compliance bottleneck nobody sufficiently planned for.
The Digital Omnibus revision (provisional agreement, May 7, 2026): The most significant 2026 update delayed the deadline for Annex III high-risk system compliance from August 2, 2026 to December 2, 2027. GPAI obligations and prohibited-practice bans were left intact and unchanged. The revision also extends prohibitions to “nudifier” AI systems effective December 2, 2026.
The practical implication: enterprise compliance timelines should anchor on December 2, 2027 for Annex III systems, while recognizing that the penalty regime under Article 99 has been live since August 2, 2025.
The United States: A Voluntary Federal Frame, and a State-by-State Patchwork
The United States has no federal AI statute. What exists instead is a layered executive framework, a single narrow federal law, and — by far the most consequential legislative body — fifty state legislatures moving faster than Congress.
Federal executive layer:
President Trump’s Executive Order 14365 (December 11, 2025), “Ensuring a National Policy Framework for Artificial Intelligence,” established the administration’s position: AI innovation must not be stifled by “excessive regulation.” It created an AI Litigation Task Force within the Department of Justice, tasked with challenging state AI laws deemed unconstitutional or preempted by federal policy. The Attorney General formalized the Task Force on January 9, 2026. As of July 2026, no lawsuits have been filed against any state AI law.
On March 20, 2026, the administration released its National Policy Framework for Artificial Intelligence, recommending that Congress preempt state AI laws that impose “undue burdens” — without yet achieving legislation to do so.
The June 2, 2026 Executive Order, “Promoting Advanced Artificial Intelligence Innovation and Security,” added voluntary cybersecurity frameworks for frontier AI deployment and reinforced the administration’s deregulatory position, per the Council on Foreign Relations.
The one federal statute: The Take It Down Act, signed May 2025, covers nonconsensual intimate images including AI-generated deepfakes. It is the only federal AI-specific law enacted in the 119th Congress.
State legislature: By mid-2026, 29 states had enacted at least one AI law, with 109 AI statutes and 28 data-center statutes on the books as of a July 6, 2026 count by New York University’s Center on Technology Policy (reported in Tech Policy Press). Fourteen states enacted companion-chatbot laws in 2026 alone — compared to two in 2025. The most common areas: generative AI regulation, algorithmic accountability, AI use in hiring, child safety, and deepfake disclosure.
California and New York have enacted transparency laws for frontier AI developers. Colorado’s SB 24-205 remains the most structurally ambitious state law, requiring risk management programs and algorithmic impact assessments for high-risk AI systems. The Trump administration’s EO singled out the Colorado law as “probably the most excessive” (AI czar David Sacks, December 2025).
Sarah Mitchell’s read: the federal non-act is itself a regulatory position. A “minimally burdensome national policy framework” is not neutral — it is a choice to let the market allocate AI risk. The question, unanswered by any executive order, is what happens when the market allocates that risk onto the people least equipped to bear it.
China: Layered Binding Rules Without a Unified Statute
China’s AI regulatory architecture is best understood as stacked layers rather than a single comprehensive law. No unified AI statute has been enacted; the State Council’s legislative work plans referenced a draft AI law in 2023 and 2024, but as of mid-2026 no official draft has been formally issued.
What exists instead is a cascade of binding administrative measures:
| Regulation | Authority | Effective | Scope |
|---|---|---|---|
| Algorithm Recommendation Provisions | CAC + 4 agencies | March 2022 | Recommendation algorithms with public opinion capacity |
| Deep Synthesis Provisions | CAC + agencies | January 2023 | Synthetic audio, video, image, text (deepfakes) |
| Generative AI Measures | CAC + 6 agencies | August 2023 | Public-facing generative AI services |
| GB 45438-2025 AI Content Labeling | SAC/SAMR | September 2025 | Mandatory implicit + explicit labeling of AI-generated content |
| AI Anthropomorphic Interaction Measures | CAC + 4 agencies | July 15, 2026 | Human-like AI companions, chatbots with emotional interaction |
Source: CAC official publications; Deep-Lex China AI Regulation Tracker (retrieved July 2026); Bird & Bird AI Regulatory Horizon Tracker China.
The Generative AI Measures require any public-facing generative AI service with “public opinion attributes or social mobilization capabilities” to file with the CAC before deployment. Content must reflect “core socialist values” and cannot undermine state power. User identity verification is mandatory. This creates a two-tier structure: base model filing (registration with substantive review) and application filing (lighter-touch registration for apps built on approved models).
The Cyberspace Administration of China published a cumulative total of 868 generative AI service registrations and 530 application filings as of April 30, 2026, based on a May 13, 2026 official announcement. Beijing alone accounted for approximately 225 of those registrations — about 30% of the national total. Shanghai had 158 as of April 21, 2026. On July 15, 2026, the CAC published the first batch of seven on-device AI service filings, including Apple Intelligence (via Alibaba/Baidu partnerships), Huawei, Xiaomi, OPPO, vivo, Samsung, and Nubia.
Enforcement is active: local Cyberspace Administrations have issued administrative penalties against non-compliant providers, ordered app delistings, and conducted special campaigns including “Liangjian Pujiang 2025” in Shanghai focusing on AI abuse. The Nanchang Cyberspace Administration imposed a penalty on a generative AI provider for labeling violations, per Bird & Bird’s tracker.
The structural constraint the numbers don’t show: Chinese AI models must avoid politically sensitive topics — Taiwan’s status, 1989 Tiananmen, Xinjiang — as a precondition for CAC approval. That ideological screen is built into the regulatory infrastructure. It limits Chinese AI products’ global competitiveness in open-information markets, even as domestic adoption scales.
The Axis Regulatory Divergence Index™ (RDI™): Scoring Each Regime
Axis Intelligence Research created the RDI™ to provide a single citable comparative score for how prescriptive and comprehensive each major AI regulatory architecture is. This is an Axis Intelligence Research proprietary metric, published here for the first time.
Formula:
RDI™ = (0.25 × Legal Score) + (0.20 × Enforcement Score) + (0.20 × Scope Score) + (0.20 × Penalty Score) + (0.15 × Speed Score)
Each component scored 0–100 based on primary-sourced inputs. Higher = more prescriptive regulation.
| Component | Weight | Basis |
|---|---|---|
| Legal Score | 25% | Binding statute vs. executive order vs. sector guidance; legislative completeness |
| Enforcement Score | 20% | Live enforcement authority; published penalties; active investigations |
| Scope Score | 20% | Sectors and AI system types covered; extraterritorial reach; definition breadth |
| Penalty Score | 20% | Maximum fine as % of global turnover; graduated structure; SME provisions |
| Speed Score | 15% | Elapsed time from first draft regulation to active enforcement |
RDI™ Component Scores and Sources
| Jurisdiction | Legal | Enforce | Scope | Penalty | Speed | RDI™ |
|---|---|---|---|---|---|---|
| European Union | 95 | 78 | 96 | 92 | 74 | 88.0 |
| China | 70 | 82 | 72 | 60 | 91 | 74.0 |
| United States | 32 | 30 | 35 | 18 | 52 | 32.4 |
EU RDI™ = 88.0. Inputs: Regulation (EU) 2024/1689 (full binding statute, four-tier structure, 14 prohibited practices); EU AI Office operational; Article 99 three-tier fines up to 7% global turnover; prohibitions live February 2025; GPAI enforcement live August 2025.
China RDI™ = 74.0. Inputs: Five layered binding administrative measures (2022–2026); CAC active enforcement with 868 registered services; content labeling mandatory September 2025; enforcement speed fastest of three (first GenAI rule effective August 2023, just four months after generative AI became a mass-market category); penalty transparency lower than EU; no extraterritorial application.
US RDI™ = 32.4. Inputs: No federal AI statute; EO 14365 December 2025 (voluntary framework, no fine structure); AI Litigation Task Force established but no suits filed; state patchwork (109 enacted laws, but fragmented and jurisdiction-limited); Take It Down Act as only federal statute; maximum federal AI penalty: none.
RDI™ is a snapshot as of July 2026. Inputs and scores will change as the Digital Omnibus is formally adopted, the US National AI Policy Framework progresses through Congress, and China’s draft unified AI law advances. The formula version is RDI™ v1.0. The methodology is disclosed above in full; any researcher can replicate the calculation from the primary sources cited.
Sarah Mitchell: the 55.6-point gap between the EU and the US is the cleanest single-number description of the strategic bet each side is making. The EU bet that comprehensive ex-ante rules build trust and reduce long-term harm. The US bet that speed and minimal friction produce the models that set the global standard. Both bets have empirical track records in other technology sectors. Neither is obviously right yet.
How Do the Fines Compare?
EU Fine Structure (Article 99, Regulation (EU) 2024/1689)
The EU AI Act establishes three penalty tiers, with fines calculated as the higher of a fixed euro amount or a percentage of worldwide annual turnover for large companies — and the lower for SMEs and startups:
| Violation Type | Fixed Cap | Revenue Cap |
|---|---|---|
| Prohibited AI practices (Article 5) | €35,000,000 | 7% global turnover |
| High-risk / GPAI non-compliance | €15,000,000 | 3% global turnover |
| Misleading information to authorities | €7,500,000 | 1% global turnover |
Source: Regulation (EU) 2024/1689, Article 99; confirmed in force since August 2, 2025.
These ceilings exceed GDPR’s maximum of €20 million or 4% of turnover — making the AI Act the highest-penalty digital regulation in EU history. Enforcement is split: national market surveillance authorities handle most cases; the EU AI Office enforces GPAI model obligations directly.
For context: a large enterprise non-compliant with high-risk AI documentation requirements (the 3%/€15M tier) faces an exposure ceiling that, for a company with $10 billion in global turnover, reaches $300 million. Compliance investment of €100,000–€500,000 for a large provider is straightforwardly rational against that number.
The Digital Omnibus left Article 99 intact. The penalty regime is live and unchanged regardless of the Annex III deadline extension.
US Fine Structure
There is no federal US AI fine structure. No executive order or federal law imposes percentage-of-revenue AI penalties. State penalties vary significantly:
- Colorado SB 24-205: civil enforcement; penalty amounts set by state AG rulemaking.
- California AI transparency statutes: civil enforcement provisions under state consumer protection law.
- Take It Down Act: criminal penalties for creating/distributing nonconsensual AI-generated intimate images.
The fragmentation is the key compliance risk: a company operating across 50 states potentially faces 50 different enforcement regimes with no federal safe harbor.
China Fine Structure
China’s AI fine framework is less transparent than the EU’s. Penalties under the Generative AI Measures reference the existing Cybersecurity Law, Personal Information Protection Law, and Internet Information Service rules. Administrative penalties include:
- Warnings and orders to rectify
- Fines (amounts set by individual regulations, typically in RMB hundreds of thousands)
- App delisting / suspension of service
- Public criticism / naming-and-shaming
The enforcement mechanism most feared by companies is service suspension — losing access to China’s 1.4 billion-person market. That threat functions as a penalty ceiling with no stated cap.
Timeline: How Each Regime Has Moved
| Date | EU | US | China |
|---|---|---|---|
| March 2022 | — | — | Algorithm Recommendation Provisions effective |
| January 2023 | — | — | Deep Synthesis Provisions effective |
| August 2023 | — | — | Generative AI Measures effective |
| October 2023 | — | Biden AI Safety EO (rescinded Jan 2025) | — |
| August 2024 | AI Act enters into force | — | — |
| February 2025 | Prohibited practices ban live | — | — |
| August 2025 | GPAI rules + Article 50 transparency live | — | AI Content Labeling (GB 45438-2025) effective |
| December 2025 | — | EO 14365 (national AI framework, state preemption) | — |
| January 2026 | — | AI Litigation Task Force established | — |
| March 2026 | — | National AI Policy Framework published | — |
| April 2026 | — | — | Anthropomorphic AI Interaction Measures issued |
| May 2026 | Digital Omnibus provisional agreement: Annex III extended to Dec 2027 | — | — |
| June 2026 | — | EO “Promoting AI Innovation and Security” | — |
| July 2026 | — | — | On-device AI filing first batch (Apple, Huawei, Xiaomi, et al.) |
| August 2026 | Annex III deadline (pending Omnibus formal adoption) | — | — |
| December 2026 | Nudifier ban effective | — | — |
| December 2027 | Annex III high-risk compliance (post-Omnibus) | — | — |
Source: EU OJ 2024; EO 14365; EO June 2, 2026; CAC official announcements; Quantamix Solutions post-Omnibus compliance guide (May 2026); Deep-Lex China tracker.
Sarah Mitchell: The timeline shows that China got there first on category-specific rules, the EU got there first on comprehensive binding law, and the US federal government hasn’t technically gotten there at all — with states filling the gap at a volume that is now difficult to track. That ordering is not coincidental. It reflects each system’s political economy: China can issue binding administrative rules quickly; the EU can achieve supranational legislative consensus; the US federal legislature cannot agree on a single framework.
What Does Compliance Actually Cost?
EU Compliance Costs
EU AI Act compliance costs depend almost entirely on which of four profiles applies (provider vs. deployer; high-risk vs. not):
| Profile | Year-One Estimate | Recurring Annual |
|---|---|---|
| SME deployer (no high-risk use cases) | €5,000–€25,000 | ~20–40% of initial |
| Mid-size deployer (≥1 high-risk candidate) | €25,000–€100,000 | ~20–40% of initial |
| Large enterprise / high-risk AI provider | €100,000–€500,000+ | ~20–40% of initial |
Source: AIActBlog.nl market analysis based on mid-2026 compliance market rates; confirmed structurally by EU Commission impact assessment methodology.
These figures are modeled estimates from compliance market pricing, not the Act’s fee schedule (which doesn’t exist). Real-world costs vary by sector, existing governance infrastructure, and whether third-party conformity assessment is required.
The compliance market created by the Act is substantial. One analysis estimated it could reach €17 billion by 2030, driven by 65,000+ high-risk AI systems requiring ongoing monitoring, documentation, and certification (Prieditis, Medium, November 2025).
A Dutch government subsidy (SLIM scheme) covers up to 60% of advisory and training costs for qualifying SMEs — a genuine structural protection, not a marketing talking point.
US Compliance Costs
With no federal AI statute, US “AI compliance” currently means:
- State-by-state assessment (45 states active in 2026)
- Sector-specific guidance from federal agencies (SEC, FTC, EEOC, HUD for lending and hiring)
- Voluntary frameworks: NIST AI Risk Management Framework (RMF 1.0)
The absence of a federal penalty ceiling removes the straightforward cost-benefit anchor that makes EU compliance investment rational. US companies instead face uncertain multi-state exposure and reputational risk.
China Compliance Costs
CAC filing involves technical documentation, content governance systems, user identity verification, and security assessments. The compliance overhead is operational (building compliant systems) rather than primarily legal (navigating judicial processes). A company that fails the approval process does not launch in China; the compliance cost is effectively the market-access barrier.
Who Does the World Trust to Regulate AI?
The Pew Research Center’s October 2025 survey of 28,333 adults across 25 countries provides the most rigorous cross-national measurement available:
- EU: 53% median trust (54% among EU member nations; 48% among non-EU nations)
- United States: 37% median trust (48% distrust)
- China: 27% median trust (60% distrust)
Source: Pew Research Center, “How People Around the World View AI,” October 15, 2025. Survey fieldwork January 8–April 26, 2025.
Among all countries surveyed by Stanford HAI’s 2026 AI Index, the United States reported the lowest level of trust in its own government to regulate AI, at just 31% — below every other nation in the sample. The EU is trusted more than either the US or China to regulate AI effectively, per both the Pew survey and the 2026 Stanford AI Index public opinion chapter.
The pattern holds across demographics with one exception: younger adults tend to express higher trust in China as an AI regulator relative to older adults, per Pew.
These numbers deserve to be read carefully. Trust in a regulator is partly a function of familiarity with the regulation and partly a function of general views toward the governing entity. The EU’s lead likely reflects both its comprehensive law and its global reputation as a consumer protection authority — not a verdict that the EU’s regulatory choices are correct.
How Regulation Shapes AI Investment and Innovation
The investment data from Stanford HAI’s 2026 AI Index (April 2026) provides the clearest available picture of how regulatory posture correlates with — though does not determine — capital allocation:
- US private AI investment (2025): $285.9 billion — 23.1 times China’s $12.4 billion
- Global corporate AI investment (2025): $581.7 billion, up 130% year-over-year
- China private AI investment (2025): $12.4 billion (likely understated; does not include state guidance funds)
Source: Stanford HAI 2026 AI Index Report, April 2026.
The investment gap is real but potentially misleading. China’s AI spending through government-directed industrial policy and state-owned enterprise is not captured in private investment figures. China’s $47.5 billion semiconductor fund alone (announced 2025, per Stanford HAI 2025 AI Index) exceeds total US government AI program spending.
The relationship between regulatory stringency and investment is not monotonic. The EU’s strict AI Act coexists with significant European AI investment growth — European VC into AI reached €17.1 billion in Q3 2025 alone, per Axis Intelligence Research’s analysis of European startup funding data.
What regulation does demonstrably affect is market access and product design. Chinese AI developers face a design constraint — no politically sensitive outputs — that limits global product competitiveness in open-information markets. EU requirements force documentation and conformity assessments that non-EU developers must address before entering European markets. US developers operating in the EU must comply with the AI Act regardless of their Washington DC address.
What AI Practices Are Banned Outright?
EU Prohibitions (Effective February 2, 2025)
Under Article 5 of Regulation (EU) 2024/1689, the following are prohibited in the EU:
- AI systems that deploy subliminal techniques to influence behavior beyond conscious awareness
- AI exploiting vulnerabilities of specific groups (age, disability, socioeconomic status)
- General-purpose social scoring by public or private entities
- Real-time biometric identification in public spaces (with narrow law enforcement exceptions)
- Emotion recognition in workplace and educational settings
- Predictive policing based on individual profiling
- Facial recognition databases assembled by scraping internet/CCTV images
Effective December 2, 2026 (per Digital Omnibus): Nudifier applications — AI that generates or manipulates sexually explicit images without consent.
China’s Content Prohibitions (Generative AI Measures, Article 9)
- Content that subverts state power or the socialist system
- Content endangering national unity or territorial integrity
- Content promoting terrorism, extremism, or violent crime
- Content discriminating based on ethnicity, belief, or religion
- Content deemed false information or spreading rumors
- Content that undermines “core socialist values”
- Content promoting pornography, gambling, drugs
These restrictions apply to publicly accessible AI services within China. They are enforced through the CAC registration process — non-compliant model outputs block approval.
US Federal Bans
There are no federal-level categorical AI bans in the United States. Some sector-specific restrictions exist:
- FTC guidance prohibits deceptive AI use in commercial contexts
- EEOC guidance restricts discriminatory algorithmic employment decisions
- Take It Down Act: specific criminal prohibition on nonconsensual AI intimate imagery
Frequently Asked Questions
What is the EU AI Act and when does it fully apply?
The EU AI Act (Regulation (EU) 2024/1689) entered into force on August 1, 2024. Its prohibited-practices ban has applied since February 2, 2025. GPAI model rules and transparency obligations have applied since August 2, 2025. Under the Digital Omnibus provisional agreement of May 7, 2026, full compliance for Annex III high-risk systems is now expected by December 2, 2027 — extended from the original August 2, 2026 deadline.
What are the maximum fines under the EU AI Act?
The EU AI Act carries three penalty tiers: up to €35 million or 7% of global annual turnover for prohibited AI practice violations (Article 5); up to €15 million or 3% for high-risk and GPAI non-compliance; and up to €7.5 million or 1% for misleading information to authorities. These exceed GDPR’s maximum of €20 million or 4% of turnover.
Does the US have a federal AI law?
No. As of July 2026, the United States has no comprehensive federal AI statute. The only federal AI-specific law enacted in the current Congress is the Take It Down Act (May 2025), which addresses nonconsensual AI-generated intimate imagery. Executive Order 14365 (December 2025) establishes policy priorities and preemption goals but does not create penalties or a regulatory authority. State legislatures have enacted 109 AI laws as of mid-2026.
How does China regulate generative AI?
China’s Interim Measures for the Management of Generative Artificial Intelligence Services (effective August 15, 2023) require public-facing generative AI services to register with the Cyberspace Administration of China, meet content standards including “core socialist values,” verify user identities, and label AI-generated outputs. As of April 30, 2026, 868 services had completed full registration and 530 applications had been filed. A separate labeling standard (GB 45438-2025) took effect September 1, 2025. Rules for AI anthropomorphic interaction services took effect July 15, 2026.
Which AI regulatory approach is trusted most globally?
Pew Research Center’s 2025 survey across 25 countries found that a median of 53% of adults trust the EU to regulate AI effectively, compared to 37% for the United States and 27% for China. The Stanford HAI 2026 AI Index found that the United States reported the lowest level of domestic trust in its own government to regulate AI among all surveyed countries, at 31%.
What is the Axis Regulatory Divergence Index™ (RDI™)?
The RDI™ is a proprietary composite metric created by Axis Intelligence Research to score and compare AI regulatory architectures across five dimensions: legislative completeness (25%), enforcement infrastructure (20%), scope breadth (20%), penalty severity (20%), and regulatory speed (15%). As of July 2026, the EU scores 88.0, China 74.0, and the United States 32.4 on a 0–100 scale. The formula and all component scores are disclosed above; the metric is reproducible from primary sources.
Can a US company be penalized under the EU AI Act?
Yes. The EU AI Act applies to any entity placing an AI system on the EU market or deploying AI systems that affect people within the EU, regardless of the company’s country of incorporation. A California-headquartered AI company whose products are used in France is subject to the Act. Non-compliance exposes it to fines up to 7% of worldwide annual turnover — including US revenues.
How many AI laws have US states enacted?
According to a July 6, 2026 count by NYU’s Center on Technology Policy (reported in Tech Policy Press), US states had enacted 109 AI laws in the first half of 2026 alone, alongside 28 data-center statutes. In 2025, lawmakers in all 50 states introduced AI-related bills for the first time; 145 were enacted into law. As of March 2026, 45 states had introduced 1,561 AI-related bills.
Methodology
Research process: All statistics in this article were collected from primary sources during a dedicated research session in July 2026. No figure originates from model training memory without source verification. The research workflow followed Axis Intelligence Research Editorial Protocol V9: fact table built before prose; verification pass run against all cited URLs.
Primary sources consulted:
- EU Official Journal, Regulation (EU) 2024/1689 (the AI Act text)
- European Commission Digital Strategy / EU AI Office official documentation
- CAC official announcements (Xinhua, CGTN, Wedoany — citing CAC statements)
- White House Executive Order 14365 (December 11, 2025)
- White House National Policy Framework for Artificial Intelligence (March 20, 2026)
- Stanford HAI 2026 AI Index Report (April 2026)
- Pew Research Center, “How People Around the World View AI” (October 15, 2025)
- MultiState AI Legislative Tracker (multistate.ai)
- NYU Center on Technology Policy / Tech Policy Press (July 6, 2026 state law count)
- Bird & Bird AI Regulatory Horizon Tracker – China
- Deep-Lex China AI Regulation Tracker
- Quantamix Solutions EU AI Act Compliance Guide (updated May 2026)
RDI™ methodology: The Regulatory Divergence Index™ is an Axis Intelligence Research proprietary composite. It is not an audit tool or legal assessment. It measures the publicly observable design of regulatory frameworks, not their enforcement effectiveness in any individual case. The five dimensions and weights were selected to capture the variables that most directly affect compliance planning for an enterprise operating internationally. Component scores are based on binary and continuous primary-source inputs. Reproducibility: any researcher can recompute from the inputs and formula above.
Limitations:
- US state law count changes daily; all figures carry the date of the cited source.
- China fine amounts are not publicly tabulated in a single official source; the penalty structure summary draws on regulatory text, not published enforcement decisions.
- RDI™ v1.0 scores will be revised when the Digital Omnibus is formally adopted, the Annex III deadline is confirmed, and if China enacts a unified AI statute.
- EU compliance cost estimates are market-based models, not audit-certified figures.
About This Dataset
The CSV accompanying this article (us-eu-china-ai-regulation.csv) is the fact table underlying every quantitative claim in this article. It contains one row per verifiable data point, with full provenance: source organization, source document, URL, retrieval date, primary source flag, Axis-calculated flag, and method note for any derived figures.
The dataset includes:
- EU AI Act penalty structure (3 tiers × 2 metrics)
- EU AI Act enforcement timeline (7 dated milestones)
- China GenAI registration counts (3 official CAC datapoints)
- US state AI legislation figures (2025 and 2026)
- Pew Research trust scores (3 jurisdictions)
- Stanford HAI AI investment figures (US and China, 2025)
- RDI™ component scores and composite outputs (Axis-calculated)
License: CC BY 4.0. You may use, share, and adapt this data with attribution: “Axis Intelligence Research, US vs EU vs China AI Regulation 2026, July 2026, axis-intelligence.com/us-eu-china-ai-regulation/”
Cite this article:
APA: Axis Intelligence Research & Mitchell, S. (2026, July). US vs EU vs China AI Regulation 2026: Laws, Fines, and Compliance Compared. Axis Intelligence Research. https://axis-intelligence.com/us-eu-china-ai-regulation/
MLA: Axis Intelligence Research and Sarah Mitchell. “US vs EU vs China AI Regulation 2026: Laws, Fines, and Compliance Compared.” Axis Intelligence Research, July 2026, axis-intelligence.com/us-eu-china-ai-regulation/.
Chicago: Axis Intelligence Research and Sarah Mitchell. “US vs EU vs China AI Regulation 2026: Laws, Fines, and Compliance Compared.” Axis Intelligence Research, July 2026. https://axis-intelligence.com/us-eu-china-ai-regulation/.
Last updated: July 19, 2026