Global AI Governance: Mapping the Divergence (Brussels vs.

Summary: As AI permeates the global economy, three distinct governance models have emerged. This guide maps the philosophical and practical divergence between the EU's "Rules-Based" model, the US's "Market-Driven" approach, and China's "State-Controlled" vertical strategy, helping enterprises navigate the fragmented compliance map.

---

For global technology firms, the dream of a unified "Global AI Law" is dead. Instead, the world has splintered into three distinct regulatory blocs, each driven by a fundamentally different primary objective.

Understanding these divergent poles is not an academic exercise—it is an operational necessity. A model that is compliant in San Francisco (US) may be illegal in Paris (EU) and require a state filing in Beijing (China).

The Three Poles of Governance

1. The Brussels Model: "Rights-Based Regulation"

Primary Goal: Protection of Fundamental Rights.
Mechanism: Comprehensive, horizontal legislation (The EU AI Act).

Philosophy: Ex-ante rules (permission first). You must prove safety before market entry.

Key Driver: The "Brussels Effect"—using market size to export standards globally.

2. The Washington Model: "Market-Driven Innovation"

Primary Goal: National Security and Economic Leadership.

Mechanism: Distributed agency enforcement (FTC, NIST) and Executive Orders.

Philosophy: Ex-post enforcement (forgiveness later). Innovate first, punish harms as they occur.

Key Driver: Competition with China.

3. The Beijing Model: "State-Centric Control"

Primary Goal: Social Stability and State Security.
Mechanism: Targeted vertical regulations (Recommendations, Deepfakes, GenAI).

Philosophy: Strict licensure*. State approval required for algorithms with public influence.

Key Driver: Regime stability and ideological control.

The Convergence (Where they Agree)

Despite the friction, there are areas of overlapping consensus, largely driven by the Bletchley Park process and G7 guidelines.

Transparency: All three blocs demand labeling of AI-generated content (Watermarking).
Red Teaming: The requirement for adversarial testing of frontier models is becoming a global standard.
Prohibited Applications: Social Scoring is explicitly banned in the EU and functionally taboo in the US, though state surveillance remains a differentiator in China.

Comparison Matrix: The Compliance Trilemma

Feature	European Union	United States	China
Legal Basis	AI Act (Hard Law)	EO 14110 / Agencies	Specific Decrees
Foundation Models	Tiered (Systemic Risk)	Reporting (>10^26 FLOPS)	Security Assessment
Liability	Strict Product Liability	Case-by-case (Torts)	Service Provider Liable
Cross-Border	Privacy Shield Constraints	"Protect our Data"	Strict Localization

Enterprise Action Plan: The Global Strategy

How do you build a product that works everywhere? You don't. You build a compliance layer that adapts.

1. Baseline on the Strictest Rule: For data governance and documentation, adopt the EU AI Act standard globally. It is the most rigorous and covers 90% of US/China documentation requirements.

2. Isolate China Operations: Due to both US export controls on chips and China's data export laws, your China AI stack must be logically air-gapped from your global stack.

3. Adopt NIST for RMF: Use the NIST AI Risk Management Framework as your internal language of risk. It is the most portable "dialect" of compliance recognized by regulators worldwide.

4. Watch the "Brussels Effect": Monitor Brazil, Canada, and Australia. They are currently cutting-and-pasting the EU model, not the US one.

To understand the specific rules of each bloc, read our deep dives on The EU AI Act, US AI Governance, and China's Algorithmic Control. For the legal battles over data, see Copyright & Generative AI.

---