AI Preemption War: Part 6 — The International Context

Series: The AI Preemption War | Table42 Research

—

When the European Union passed the AI Act in 2024, the United States had a choice: develop a comparable framework, defer to the EU model, or take a different path. The National Policy Framework takes the third option—not by designing a distinct American approach, but by deferring most questions and preempting state alternatives.

The result is a US AI policy that stands apart from its major competitors: more permissive than the EU and less intentional than China’s state-directed approach. In user-facing individual rights—data privacy, algorithmic transparency, content labeling—the US framework offers fewer protections than either model. But the comparison with China requires care: China’s regulatory requirements operate within a system designed primarily for state control and social management, not individual empowerment. China’s “protections” are instruments of governance, not constraints on government power. The meaningful comparison is with the EU, where individual rights protections are designed to constrain both companies and the state.

The EU AI Act as Counter-Model

The EU AI Act represents the most comprehensive AI governance framework adopted by any major democracy. It classifies AI systems by risk level—unacceptable risk (prohibited), high risk (strict requirements), limited risk (transparency obligations), minimal risk (voluntary standards)—and imposes corresponding obligations.

Unacceptable risk: The EU Act prohibits AI systems that manipulate people through subliminal techniques, exploit vulnerabilities, enable social scoring by governments, or conduct real-time biometric surveillance in public spaces (with narrow exceptions). These systems are banned outright.
High risk: AI systems used in employment decisions, credit decisions, education assessments, criminal justice, and critical infrastructure must meet conformity assessment requirements: extensive documentation, risk management systems, data governance requirements, transparency obligations, human oversight mechanisms, and accuracy standards. These systems cannot be deployed in the EU without meeting requirements and passing assessment.
Limited risk: AI systems that interact directly with people—chatbots, emotion recognition systems, deepfake generators—must disclose AI nature to users. This is the transparency regime the US framework mentions but doesn’t mandate.
Minimal risk: AI systems in low-risk contexts (spam filters, AI games) face no mandatory requirements; the EU encourages voluntary codes of conduct.

The EU Act also establishes enforcement mechanisms: national market surveillance authorities, substantial penalties (up to €35 million or 7% of global turnover for prohibited-practice violations; lower tiers face up to 3% or 1.5%), and a new European AI Office to coordinate enforcement. As of March 2026, however, only 8 of 27 EU member states have formally designated their competent authorities as required by Article 70—meaning the enforcement architecture remains incomplete more than a year after the August 2, 2025 statutory deadline.

What the EU Act Hasn’t Solved

The EU AI Act is a statute, not a working regulatory system—yet. As of early 2026, several EU member states have not designated their competent national authorities. The European AI Office is operational but understaffed. High-risk system conformity assessments—the Act’s core enforcement mechanism—require technical standards that are still being developed by CEN and CENELEC, the EU’s standards bodies. The prohibited-practices provisions took effect in February 2025, but the high-risk requirements don’t fully apply until 2027.

European AI companies have been vocal critics. France’s Mistral and Germany’s Aleph Alpha have argued publicly that the Act’s compliance costs drive AI investment toward the US, where no equivalent framework exists. A 2025 study by the Centre for European Policy Research found that EU-based AI startups were 23% more likely to establish a US subsidiary than non-AI EU startups—a gap the authors attributed partly to regulatory burden. The Brussels Effect cuts both ways: EU regulation may set global standards, but it may also export European AI talent and capital to less-regulated jurisdictions.

The April 2026 negotiations collapse. Trilogue negotiations to amend the EU AI Act under the “Digital Omnibus for AI” stalled in April 2026 after Parliament and Council failed to agree on amendments. The IAPP reported that Ashley Casovan, AI Governance Center Managing Director, described the delay as significant for companies awaiting regulatory clarity. Forty-plus European companies—including ASML and Mistral—publicly urged a two-year implementation delay, arguing the current requirements put the bloc’s AI ambitions at risk. The European Commission has stated it has “no plans to pause” implementation, but the failure to amend creates uncertainty about what the final regulatory landscape will look like. The EU AI Act remains a statutory framework with unresolved implementation—not yet the functioning regime the series’ earlier framing implied.

The EU Act’s risk classification system has also been criticized by legal scholars as incoherent in places—some systems straddle multiple risk categories, and the line between “high risk” and “limited risk” applications can depend on deployment context rather than system design. None of this means the EU approach is wrong. But it means the comparison should be US reality vs. EU reality, not US aspirations vs. EU statutes.

The Chinese Model

China’s approach to AI governance reflects the country’s distinct political structure: state-directed development, limited civil society, and the government’s interest in maintaining social control.

China has issued regulations on generative AI services, algorithmic recommendations, deep synthesis (deepfakes), and generative AI content. These regulations impose requirements on AI companies operating in China—including transparency requirements, content labeling, and government access provisions—but the primary orientation is toward government control rather than user protection.

Chinese AI regulation also serves industrial policy goals: the government directs AI development toward areas it considers strategically important, uses regulatory requirements to shape industry behavior, and balances control with promotion of domestic AI champions.

The US framework does not reference the Chinese model directly, but the “competitiveness” framing implies comparison. The argument is that the US needs lighter regulation than the EU to maintain AI competitiveness relative to China.

What the US Approach Actually Is

The US framework has no comprehensive AI statute, no dedicated AI regulator, and no enforcement mechanism of the scale the EU has created. Its primary tool for shaping AI governance is preemption: displacing state laws to create national uniformity.

This approach has real consequences:

For AI companies: National uniformity reduces compliance costs—they don’t need to navigate different state requirements. The “burdensome” standard for preemption means companies can argue that detailed requirements are preempted. The limited liability language reduces exposure.
For state governments: Preemption displaces AI laws they enacted in response to constituent demands. State legislators who passed transparency, anti-discrimination, and consumer protection laws see their work undone.
For residents: Federal minimum standards—if they exist at all—replace state protections. No federal AI transparency law, no federal anti-discrimination requirements for AI, no federal consumer protection for AI harms.
For democratic accountability: National policy is made by federal officials, influenced by national-level lobbying and industry pressure. State policy reflected state-level preferences and accountability. Preemption shifts the locus of AI governance toward federal institutions and away from state ones.

The Missing Middle

What the US approach lacks is a middle ground between comprehensive EU-style regulation and pure preemption. Options that don’t appear in the framework:

Federal baseline with state opt-up: Establish minimum federal standards that states can exceed. This preserves state experimentation while preventing a race to the bottom.
Liability calibration: Rather than warning against “open-ended liability,” establish clear standards that match liability to harm and create incentives for precaution without eliminating claims entirely.
Sector-specific frameworks: Rather than a general preemption approach, develop specific rules for specific domains—hiring AI, healthcare AI, financial AI—through existing regulatory agencies with defined enforcement mechanisms.
Transparency mandates: Require disclosure of AI-generated content, algorithmic decision-making, and training data sources without comprehensive prohibition or liability limitation.

The framework doesn’t take any of these approaches. It defers on copyright, warns against liability, proposes preemption of state law, and relies on industry self-regulation for most governance questions.

The International Coordination Problem

US AI policy doesn’t exist in isolation. AI companies operate globally; AI systems affect users in multiple jurisdictions; AI governance questions are being worked out in international forums.

The EU AI Act includes extraterritorial provisions: companies offering AI services in the EU must comply regardless of where they are based. This creates pressure on US companies to meet EU standards even without US legal requirements.

The framework’s deferral on international questions leaves US companies navigating a patchwork: EU requirements where they serve EU users, state requirements where they serve US users, and no comprehensive US framework to coordinate these demands.

International AI governance institutions are developing—standards bodies, bilateral agreements, multilateral forums. The US approach to these institutions will shape global AI norms. The framework provides no guidance on what those norms should be or how the US should engage.

What This Series Has Shown

The AI Preemption War series has traced a structural contest: states that developed AI governance to protect residents, a federal framework that would displace those protections, an industry that benefits from uniformity, and a policy debate framed in terms of competitiveness and innovation that obscures the interests at stake.

The framework is not inevitable. Congress can choose different approaches. State resistance—political, legal, administrative—is ongoing. Courts will face preemption questions when legislation passes. The administration has taken positions that future administrations could reverse.

But the direction is contested, not set. Congressional opposition remains significant, state resistance is organized, and the framework’s non-binding status means nothing is certain until legislation passes. What is clear is the trajectory: US AI governance is being pushed toward preemption and away from state-level accountability. Whether that push succeeds—and what it means for residents, workers, consumers, and democracy—is the architectural question Table42 will keep examining.

Primary Sources

US Federal Framework

1. National Policy Framework for Artificial Intelligence (March 20, 2026): Preemption language, innovation framing
2. Executive Order 14365 (December 11, 2025): Foundation for national AI policy
3. National Artificial Intelligence Initiative Act of 2020, 15 U.S.C. § 9411 et seq.: Federal AI coordination
4. CHIPS and Science Act of 2022, 15 U.S.C. § 4651 et seq.: AI investment provisions

EU AI Act — Comprehensive Framework

5. EU AI Act (Regulation 2024/1689): Full text — risk classification system
6. EU AI Act Article 5: Prohibited AI practices — unacceptable risk systems
7. EU AI Act Articles 6-7: High-risk AI classification criteria
8. EU AI Act Articles 8-15: Requirements for high-risk AI — transparency, accuracy, human oversight
9. EU AI Act Articles 50-52: Limited risk — transparency obligations
10. EU AI Act Articles 70-77: Enforcement mechanisms and penalties
11. EU AI Act Article 53: Training data transparency for general-purpose AI
12. European AI Office: Implementation body — organizational structure and mandate
13. European Commission: AI Act implementation guidelines (2025)
14. EU Member State competent authorities: National enforcement designations

Chinese AI Governance

15. China Interim Measures for the Management of Generative AI Services (August 2023): Generative AI content requirements
16. China Provisions on the Management of Algorithmic Recommendations (March 2022): Algorithm transparency and content rules
17. China Deep Synthesis Provisions (January 2023): Deepfake regulation and labeling
18. China Cybersecurity Law (2017): Data localization and security requirements
19. China Personal Information Protection Law (2021): Data protection framework
20. China State Council: New Generation Artificial Intelligence Development Plan (2017)
21. China Ministry of Science and Technology: AI governance guidelines (2024)
22. US-China Economic and Security Review Commission: Chinese AI development report (2025)

International AI Governance

23. OECD AI Principles (2019): International AI governance guidelines — 46 member countries
24. OECD AI Policy Observatory: Cross-country regulatory comparison database
25. G7 Hiroshima AI Process (2023): International AI governance framework
26. G7 Hiroshima AI Code of Conduct (2023): Voluntary AI developer guidelines
27. Bletchley Declaration (November 2023): AI Safety Summit communique — UK-hosted
28. Seoul AI Safety Summit (May 2024): Follow-up international agreement
29. United Nations AI Advisory Body: Recommendations on global AI governance (2024)
30. UNESCO Recommendation on the Ethics of AI (2021): International ethical framework
31. Council of Europe: AI and human rights framework (2024)
32. GPAI (Global Partnership on AI): Multi-stakeholder AI governance initiative

Extraterritoriality and Compliance

33. EU AI Act Article 2: Territorial scope — extraterritorial application to non-EU providers
34. EU General Data Protection Regulation (GDPR) Article 3: Extraterritoriality precedent
35. California Consumer Privacy Act (CCPA) § 1798.140: Extraterritorial application to out-of-state businesses
36. Brussels Effect: Anu Bradford scholarship on EU regulatory extraterritoriality
37. International trade implications: WTO and AI governance compatibility

Comparative Analysis

38. Stanford HAI: “AI Index Report 2025” — global AI policy comparison
39. Center for Strategic and International Studies: “AI Governance: US vs. EU vs. China” (2025)
40. Brookings Institution: “Divergent Paths: AI Regulation in US and EU” (2024)
41. Carnegie Endowment for International Peace: “AI Governance Without Global Standards” (2025)
42. Information Technology and Innovation Foundation: “EU AI Act vs. US Innovation Approach” (2024)
43. European Center for Law and Justice: AI and fundamental rights analysis
44. Future of Life Institute: International AI safety coordination proposals
45. Center for AI Safety: International AI safety standards development

State Laws at Risk

46. California AB 2013, SB 942 (2024): AI transparency requirements
47. Colorado SB 24-205 (2024): Algorithmic discrimination framework
48. Illinois BIPA (2008), AI Video Interview Act (2020): Biometric and hiring AI protections
49. New York Local Law 144 (2023): Employment AI bias audits
50. Texas TRAIGA HB 149 (2025): Comprehensive AI governance

—

Series Summary

Total: ~53,000 words | Citations: 256 (Part 1: 34, Part 2: 40, Part 3: 42, Part 4: 45, Part 5: 45, Part 6: 50) | Status: Draft series complete

—

Status: Draft | Citation count: 50 | Series complete

Share on Social Media