Last reviewed: April 9, 2026
Jurisdictions covered: China (primary), with EU comparison
Reading time: 22 minutes
This is a living page. We update it as new rules are issued. China publishes AI regulations faster than any other country.
Language note: All Chinese regulations in this article were verified against the original Mandarin text on official government websites (cac.gov.cn, tc260.org.cn, gov.cn). Translations are cross-referenced with DigiChina (Stanford) and China Law Translate where available.
China AI Regulation 2026: Every Law, Rule, and Draft in One Place
On February 12, 2026, China’s Cyberspace Administration penalized 13,421 online accounts and removed 543,000 pieces of content for failing to label AI-generated material. The legal basis was not a single AI law. It was a combination of the 2025 AI Content Labeling Measures, the 2023 Deep Synthesis Rules, and mandatory national standard GB 45438-2025 — three instruments from three different years, enforced through a coordinated campaign.
This is how China regulates AI. Not through one law, but through layers.
While the EU built a single regulation with 113 articles and spent three years debating it, China issued targeted rules one by one — algorithmic recommendations in 2022, deepfakes in 2023, generative AI in 2023, content labeling in 2025, facial recognition in 2025 — each addressing a specific technology. Justice Minister He Rong confirmed in March 2026 that a comprehensive national AI law is under “legislative research,” but no draft exists and the earliest plausible enactment is 2027 (Xinhua, March 12, 2026).
For compliance teams, this layered approach creates a different kind of challenge. The EU asks: is your system high-risk? China asks: does your system recommend content? Generate text? Use facial recognition? Collect personal data? Each question leads to a different regulation, issued by a different regulator, with different filing requirements. Miss one layer and you are non-compliant — even if you follow all the others.
This article consolidates every active AI rule, every pending draft, and every enforcement signal into one English-language reference. If you operate AI products in or for the Chinese market, this is the page you bookmark.
Key Takeaways
- China has no single AI law. It regulates through a layered system of sector-specific rules issued by multiple agencies. Three AI-specific regulations (2022-2023) sit beneath three overarching laws (Cybersecurity Law, PIPL, Data Security Law).
- The CAC is the dominant AI regulator — handling algorithm filing, generative AI measures, content labeling, and deepfake rules. But MIIT, SAMR, MOST, and MPS all have overlapping authority. A single AI product may need to satisfy five agencies.
- Filing before launch is mandatory. 796 generative AI services have filed with the CAC as of February 2026 (CAC, March 17, 2026). You cannot legally offer a public-facing AI service without completing the filing process.
- Enforcement is real and accelerating. The Didi fine (RMB 8.026 billion, 2022), the Alibaba antitrust penalty (RMB 18.228 billion, 2021), and the February 2026 labeling crackdown show that China enforces AI rules at scale.
- Three major new rules in April 2026 alone: mandatory AI ethics review (10 agencies), interactive AI service rules (AI companions), and digital human measures. The regulatory pace is not slowing down.
Every Active AI Rule in One Table
This table includes every binding rule with direct AI obligations, current as of April 9, 2026.
| Rule | Issuing Body | Effective | Scope | Key Obligation | Penalty Range |
|---|---|---|---|---|---|
| Algorithmic Recommendation Rules | CAC + MIIT + MPS + SAMR | Mar 1, 2022 | Recommendation algorithms (content feeds, search, ads, pricing) | Algorithm filing; user opt-out right; price discrimination ban; minors protection | Up to RMB 100K; suspension |
| Deep Synthesis Rules | CAC + MIIT | Jan 10, 2023 | Deepfake/synthetic media (text, image, audio, video, virtual scenes) | Mandatory watermarking/labeling; real-name registration; biometric consent; filing | Fines, suspension; criminal referral |
| Generative AI Measures (Interim) | CAC + 7 agencies | Aug 15, 2023 | Public-facing generative AI services in China | CAC security assessment + filing before launch; training data compliance; content restrictions; annual assessment | Fines, suspension; criminal referral |
| AI Content Labeling Measures | CAC + MIIT + MPS + SAMR | Sep 1, 2025 | All AI-generated content (text, image, audio, video) | Explicit + implicit labeling; metadata requirements; platform verification | Account shutdowns; content removal (13,421 accounts penalized Feb 2026) |
| Facial Recognition Measures | CAC + MPS | Jun 1, 2025 | All facial recognition processing in China (R&D exempt) | Separate consent; on-device storage; alternatives required; registration at 100K persons | Existing law penalties + criminal |
| PIPL (AI provisions) | NPC Standing Committee | Nov 1, 2021 | All personal data processing including AI | Art. 24: automated decision transparency + right to explanation + non-discrimination | Up to RMB 50M or 5% revenue |
| Cybersecurity Law (amended) | NPC Standing Committee | Jan 1, 2026 (amendment) | All network operators including AI services | AI development safety clause; data localization; security reviews | Up to RMB 10M (post-amendment) |
| Data Security Law | NPC Standing Committee | Sep 1, 2021 | All data processing including AI training data | Data classification; “important data” restrictions; cross-border controls | Up to RMB 10M + criminal |
| GB 45438-2025 (mandatory standard) | TC260 / SAC | Sep 1, 2025 | All AI-generated content | Technical labeling specifications: watermark dimensions, metadata format, audio marking | Enforceable as part of labeling measures |
| SAMR Anti-Monopoly Guidelines | SAMR | Feb 13, 2026 | Internet platforms using algorithms | Ban on algorithmic differential pricing; eased burden of proof for plaintiffs | Antitrust penalties |
| 10-Agency AI Ethics Review | MIIT + 9 agencies | Apr 2, 2026 | All AI R&D and applications | Mandatory ethics review: 6 areas (welfare, fairness, controllability, transparency, accountability, privacy) | Under existing law authority |
Not included above: Provincial regulations (Shenzhen 2022, Shanghai 2022), policy documents (AI Plus Initiative, NGAIDP 2017), and draft measures still in comment period.
What Changed in 2025-2026
The last 18 months have been China’s most active period for AI rulemaking. Here is every significant development, chronologically.
| Date | Development | Significance |
|---|---|---|
| Mar 14, 2025 | AI Content Labeling Measures issued (4 agencies) | Expanded Deep Synthesis labeling to ALL AI content |
| Mar 28, 2025 | GB/T 45392-2025 (automated decision-making security) published | First Chinese standard implementing PIPL Art. 24 |
| Mar 28, 2025 | GB/T 45652-2025 (GenAI training data security) published | Companion to national GenAI safety standard |
| Jun 1, 2025 | Facial Recognition Measures take effect | On-device storage mandate; 100K registration threshold |
| Aug 26, 2025 | AI Plus Initiative published | State Council directive on cross-sector AI integration |
| Sep 1, 2025 | GB 45438-2025 (mandatory labeling standard) takes effect | Exact watermark specs become law-equivalent |
| Sep 2025 | AI Safety Governance Framework v2.0 published | First risk-management governance methodology |
| Oct 2025 | Cybersecurity Law Amendment passed | AI safety clause embedded in foundational cyber law |
| Jan 1, 2026 | Cybersecurity Law Amendment takes effect | Higher penalties (up to RMB 10M); extraterritorial enforcement |
| Jan 30, 2026 | TC260 announces 48-standard 2026 work plan | AI security, data security, cloud computing standards |
| Feb 6, 2026 | SAMR publishes 5 AI unfair competition cases | DeepSeek brand abuse, ChatGPT impersonation, AI trade secrets |
| Feb 12, 2026 | CAC enforces AI labeling rules at scale | 13,421 accounts penalized; 543,000+ content pieces removed |
| Feb 13, 2026 | SAMR Anti-Monopoly Guidelines finalized | Algorithmic differential pricing banned; eased burden of proof |
| Mar 10, 2026 | Qinglang 2026 campaign results | 39,000+ accounts processed; 708,000+ AI content pieces removed |
| Late Mar 2026 | CAC draft: Interactive AI Services rules | AI companions, chatbots — 2-hour limits, self-harm safeguards |
| Apr 1, 2026 | TC260 draft: OpenClaw AI Agent Security Guidelines | First standards-body guidance targeting AI agents |
| Apr 2, 2026 | 10-Agency AI Ethics Review Measures issued | First mandatory cross-ministry AI ethics review framework |
| Apr 3, 2026 | CAC draft: Digital Humans rules | Mandatory labeling, consent for likeness, minors protection |
Three rules in April alone. The pace is not slowing down.
Enforcement: What the CAC Actually Does
China enforces AI rules differently from the EU or US. Instead of formal administrative proceedings against individual companies, the CAC operates through campaigns — coordinated enforcement sweeps with names like “Clear and Bright” (清朗) that target specific violation categories across hundreds or thousands of companies simultaneously.
Recent enforcement patterns:
- Feb 2026: 13,421 accounts penalized, 543,000+ content pieces removed for AI labeling violations (TechNode, Feb 13, 2026)
- Mar 2026: Qinglang 2026 Spring Festival action: 39,000+ accounts, 708,000+ content pieces targeting AI-generated “digital slop”
- Feb 2026: SAMR published 5 AI unfair competition cases — DeepSeek brand squatting (RMB 5,000-30,000 fines), fake ChatGPT accounts (RMB 62,692), AI voice robot fraud (RMB 200,000), algorithm trade secret theft (RMB 360,000)
- Apr-Jul 2025: “Clear and Bright: AI Technology Abuse Rectification” campaign targeting deepfake misuse and AI spam
The largest single-company penalties remain the 2021-2022 antitrust and data cases: Alibaba (RMB 18.228 billion), Didi (RMB 8.026 billion), Meituan (RMB 3.442 billion). These used overarching laws (PIPL, Anti-Monopoly Law), not AI-specific regulations. But they signal the enforcement apparatus’s capacity: when China decides to penalize an AI company, the fines are among the largest in the world.
For the complete enforcement case database, see our upcoming China AI Enforcement tracker.
How China Compares to the EU
| Dimension | China | EU |
|---|---|---|
| Structure | Layered: 6+ separate regulations + 3 overarching laws | Single regulation: AI Act (113 articles + annexes) |
| Classification | By technology function (recommendations, deepfakes, GenAI, facial recognition) | By risk level (unacceptable, high, limited, minimal) |
| Pre-market gate | CAC security assessment + algorithm filing | Conformity assessment (self or third-party) |
| Filing system | 796 GenAI filings + 6,000+ algorithm filings (mandatory) | EU database (not yet operational as of April 2026) |
| Enforcement style | Campaign-based (Clear and Bright); thousands of entities per sweep | Authority-driven; individual proceedings (no enforcement yet) |
| Max penalty | RMB 50M or 5% revenue (PIPL); RMB 18B+ (antitrust) | EUR 35M or 7% global turnover |
| Content restrictions | Socialist core values; explicit political content requirements | No political content requirements |
| Timeline | Rules issued in months; standards updated annually | Single law took 3 years to negotiate |
| National AI law | Under “legislative research” — 2027+ | AI Act in force since August 2024 |
| Extraterritorial reach | Art. 20 GenAI Measures: CAC can block foreign services | Art. 2: applies to systems placed on EU market |
| Harmonised standards | TC260 standards: 48 new proposed for 2026 | CEN-CENELEC: zero published, prEN 18286 failed vote |
For the full structural comparison with dual-compliance guidance, see our upcoming China vs EU AI Act analysis.
For how South Korea’s AI Basic Act fits into this picture, see our Korea cluster.
Disclaimer: This content is for informational purposes only and does not constitute legal advice. Chinese AI regulation changes frequently and operates across multiple agencies. Organizations should engage qualified legal counsel with China-specific expertise for compliance guidance. Reg Intel is not a law firm and does not provide legal services.
Last verified: April 9, 2026
Compare: EU vs China
For the global keystone comparison across twelve dimensions — algorithm filing vs conformity assessment, content moderation conflicts, asymmetric extraterritoriality, enforcement philosophy, and a five-step dual-market compliance baseline — see EU vs China AI Regulation: Two Systems, Two Philosophies (2026).