China vs the EU — and China vs the world
Lesson 5 of 5 in China: The World’s Most Developed Binding AI Rulebook.
China does not only regulate at home; it exports a governance narrative. The Global AI Governance Initiative (October 2023, announced at the Belt and Road Forum) frames AI governance around sovereignty, development rights, and opposition to “technological hegemony” — a deliberate alternative to the Western rights-based framing. It was followed by the Global AI Governance Action Plan at the World AI Conference in Shanghai (July 2025) and a proposal for a World AI Cooperation Organization headquartered in Shanghai. At the UN, China led the July 2024 General Assembly resolution on AI capacity building — adopted by consensus, and pitched squarely at the Global South, mirroring the US-led March 2024 resolution.
The audience is the majority of countries that see AI governance primarily as a development question. Whether you call the resulting pull a Beijing effect — China’s regulatory patterns and cheap open-weight models shaping choices across developing markets, the way the Brussels effect describes EU influence — is one of the live analytical debates of the decade.
Where is the comprehensive “AI Law of China”?
A draft comprehensive AI Law appeared in the State Council’s legislative plans in 2023 and 2024, and scholars (notably a CASS group) published influential model drafts. Yet Beijing keeps choosing targeted regulations instead. The revealed preference is informative: iteration preserves regulatory agility and avoids freezing rules around a moving technology — the mirror image of the EU’s bet on one comprehensive act. As of September 2026 no comprehensive AI statute has been enacted; treat any claim that “China’s AI Law” exists as a red flag, and check current status before relying on it.
Does China’s rulebook reach foreign companies?
Mostly through the market, not through extraterritorial text. The rules govern services offered to the public in mainland China — so a US model provider with no Chinese service faces nothing, while any firm entering the market inherits the full gauntlet: filing, assessment, labeling, content moderation. Contrast the EU AI Act’s explicit extraterritorial reach (outputs used in the EU) and Korea’s domestic-representative requirement. In practice, few Western consumer AI services operate in China at all — the compliance bar and the content rules are why.
Is it all just censorship?
No — and treating it that way will make you a bad analyst. The content-control provisions are real and central. But the opt-out rights, price-discrimination ban, gig-worker protections, deepfake-consent rule, and labeling architecture are genuine governance innovations, several of which prefigured Western equivalents. The professional stance: read each provision for its function — control, protection, or industrial policy — rather than assigning the whole rulebook one motive.
| Dimension | China | European Union |
|---|---|---|
Architecture | Vertical & iterative: one binding rule per technology (algorithms, deep synthesis, GenAI, labeling), revised frequently | Horizontal: one comprehensive act covering all AI by risk tier, plus a GPAI chapter |
Normative anchor | State security, content control, social stability — plus consumer protection and industrial policy | Fundamental rights, safety, and the internal market |
Speed | GenAI rules in force August 2023, about nine months after ChatGPT | AI Act entered into force August 2024; GPAI duties from August 2025; high-risk duties phasing to 2026–27 |
Market entry | Ex-ante permission in practice: security assessment + algorithm filing before public launch | Ex-ante conformity assessment for high-risk systems; most AI faces no pre-market gate |
Content of outputs | Regulated directly: prohibited content categories, “core socialist values”, moderation duties | Largely unregulated as content; the Act regulates risk, transparency, and process, not viewpoints |
Extraterritorial reach | Limited: applies to services offered to the Chinese public; leverage comes from market access | Explicit: providers abroad are covered when systems or their outputs are used in the EU |
Penalty style | Rectification orders, suspension, app-store removal, fines under existing statutes; public campaigns | Administrative fines up to 7% of global turnover for prohibited practices; national authorities + AI Office |
Standards linkage | TC260 “voluntary” standards operate as de facto mandatory assessment benchmarks | Harmonised standards (CEN/CENELEC JTC 21) grant a presumption of conformity — voluntary but strongly incentivised |
Key terms: beijing effect, Brussels effect, global ai governance initiative, extraterritoriality
Interactive checkpoint quiz (1 questions) — open this page in a browser to take it.