The vertical playbook: how China regulates AI

Lesson 1 of 5 in China: The World’s Most Developed Binding AI Rulebook.

Ask a European lawyer to describe AI regulation and they will describe one giant statute. Ask a Chinese compliance officer and they will hand you a stack of short, targeted rules — one for recommendation algorithms, one for deepfakes, one for generative AI, one for labeling synthetic content — each written fast, enforced immediately, and revised as the technology moves.

That is the deliberate design. While the EU spent 2021–2024 negotiating one horizontal law, China shipped binding rules for each technology as it emerged, making it the first country in the world to enforce dedicated rules on recommendation algorithms (2022), deepfakes (2023), and generative AI services (2023). Scholars call the approach vertical and iterative: narrow scope, quick issuance, frequent supplements. It trades the EU’s legal coherence for speed — and it means Chinese regulators had years of enforcement experience with generative AI before most jurisdictions had a single applicable rule.

Three motives drive the rulebook, and you can trace each one through every regulation you will read in this module:

  1. State security and content control. Nearly every rule polices what AI-generated content may say. The recurring statutory phrase is that services must uphold core socialist values and must not generate content endangering national security or social stability. Chinese AI regulation grew out of information control law — the Cybersecurity Law lineage — not out of product-safety or fundamental-rights law.
  2. Consumer and worker protection. Genuine, enforceable rights sit alongside the censorship provisions: opt-outs from personalised recommendation, bans on algorithmic price discrimination, protections for gig workers and minors. Dismissing the rulebook as only censorship misreads it.
  3. Industrial policy. The 2017 New Generation AI Development Plan set the goal of global AI leadership by 2030. Regulation is calibrated to discipline the sector without strangling it — which is why the generative-AI rules were softened between draft and final text, and why enterprise-internal AI is largely carved out.

The conductor of this orchestra is the Cyberspace Administration of China (CAC) — the internet regulator that leads drafting and enforcement, co-issuing each rule with ministries like MIIT (industry), MPS (public security), and, for specific rules, SAMR (market regulation) or NRTA (broadcasting).

Key terms: cac, vertical regulation, core socialist values, algorithm filing, new generation ai plan

China’s rule-by-rule buildout

  • 2022-03-01 — China’s Algorithmic Recommendation Provisions take effect:

    Binding rules — filing with the CAC, transparency, opt-outs — while the West still debates frameworks. China chooses vertical, binding, and fast.

  • 2023-01-10 — China’s Deep Synthesis Provisions take effect:

    Deepfake labeling and consent requirements — the world’s first binding synthetic-media rules, two years before the EU’s equivalent bites.

  • 2023-08-15 — China’s Interim GenAI Measures take effect:

    The first binding national generative-AI regulation: content controls, labeling, security assessments for public-facing services.

  • 2025-09-01 — China’s AI content labeling measures take effect:

    Explicit and implicit (metadata) labels required for AI-generated content across platforms — synthetic-media law at national scale.

Two more foundation layers matter before the AI-specific rules. The Personal Information Protection Law and Data Security Law (both 2021) supply the data substrate — consent, cross-border transfer controls, and security duties that every AI system in China inherits, much as the GDPR does in Europe. And the Science and Technology Ethics Review Measures (2023) require ethics review committees for research in sensitive areas, including algorithms with strong social-mobilisation potential — an upstream gate the EU has no direct equivalent for.

Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.