- China governs AI through a layered system of sectoral regulations rather than a single comprehensive law, covering algorithms, deepfakes, generative AI, and content labeling through separate but interlocking rules.
- The Cyberspace Administration of China (CAC) serves as the primary filing authority, requiring registration for algorithms with public mobilization capabilities and for all large language models.
- Mandatory AI content labeling rules took effect September 1, 2025, requiring visible labels on chatbots, AI-written content, synthetic voices, and deepfakes.
- The AI Plus Action Plan targets 70% AI penetration in key sectors by 2027 and 90% by 2030, creating parallel pressure to adopt AI while complying with detailed governance requirements.
What Happened
China has built the most layered AI governance framework of any major economy. Rather than passing a single comprehensive AI law — an approach the government removed from its 2025 legislative agenda — Beijing has issued a series of targeted regulations, each addressing a specific AI risk category. The result is a framework that covers algorithm recommendation, deep synthesis, generative AI, content labeling, and ethics review through separate but interlocking rules administered by multiple regulatory bodies.
On October 28, 2025, China’s top legislature passed major amendments to the Cybersecurity Law (CSL), bringing AI explicitly into national law for the first time. These amendments took effect January 1, 2026, and commit the government to supporting AI research and development while establishing formal requirements for AI risk assessment and security governance. The CSL amendments also direct the government to promote the construction of training data resources and computing power infrastructure.
Why It Matters
China’s approach differs fundamentally from the European Union’s AI Act, which takes a risk-tiered but unified legislative approach. Beijing has instead opted for what regulators describe as “agile and adaptive” governance — issuing narrower rules that can be updated independently as the technology evolves. This avoids the multi-year legislative cycles that have slowed EU enforcement but creates a more complex compliance landscape for companies operating in China, where obligations are spread across multiple regulatory instruments.
The framework matters beyond China’s borders. Any company deploying AI products in the Chinese market must navigate these overlapping requirements. China’s sectoral approach is also influencing regulatory thinking across Southeast Asia and the Middle East, where several governments are studying Beijing’s model as an alternative to the EU’s comprehensive statute approach.
Technical Details
The core regulatory stack includes four main pillars. The Algorithm Recommendation Provisions, effective since March 1, 2022, require disclosure of algorithmic use with user opt-out options and prohibit discriminatory pricing based on personal characteristics. Services with “public opinion attributes” must complete risk assessments and file with the CAC. The Deep Synthesis Provisions, effective January 2023, mandate labeling of synthetically generated content and require user registration, algorithm review, and content monitoring systems.
The Interim Measures for Generative AI Services, effective August 15, 2023, require providers to ensure legality of training data, conduct content monitoring, obtain personal data consent, and file all large language models with the CAC in addition to standard algorithm filings. Foreign investment in generative AI is permitted under China’s investment laws, but providers must comply with all filing and monitoring requirements.
The AI Content Labeling Measures, effective September 1, 2025, under technical standard GB45438-2025 require visible labels for chatbots, AI-written content, synthetic voices, and deepfakes. Downloadable AI content must carry embedded explicit labels. Other AI-generated content requires implicit labels such as watermarks with metadata. Platforms bear responsibility for alerting users to suspected AI-generated content.
Who’s Affected
Any company deploying AI in China faces compliance obligations across multiple regulatory instruments simultaneously. Non-compliance carries escalating penalties including regulatory investigations, administrative fines, business suspension, permit revocation, and potential criminal liability under the Cybersecurity Law, Data Security Law, and Personal Information Protection Law. The CAC has approved thousands of algorithm filings as of October 2025, indicating active enforcement of the registration requirements.
The State Council’s AI Plus Action Plan sets ambitious adoption targets: 70% AI penetration in key sectors by 2027 and 90% by 2030, with full AI-powered economy status envisioned by 2035. This creates parallel pressure for companies to adopt AI quickly while simultaneously complying with the detailed and expanding governance requirements. The tension between rapid adoption targets and regulatory compliance is a defining feature of China’s AI landscape.
What’s Next
The National Data Administration has indicated that more than 30 new standards relating to public data, data infrastructure, AI agents, high-quality datasets, and important data catalogs for various sectors will be issued in 2026. China is also developing draft ethics review measures that would extend mandatory ethical review beyond biological and medical AI to all research and development activities affecting health, safety, reputation, environment, and public order. The regulatory framework will continue expanding incrementally rather than through a single omnibus AI statute, meaning compliance requirements will grow with each new sectoral rule.
Related Reading
- Singapore Leads on Agentic AI Governance With World-First Framework
- White House Releases Federal AI Framework, Urges Congress Against Heavy Regulation
- US Advisory Body Warns China Open-Source AI Strategy Creates Self-Reinforcing Advantage Over American Labs
- US AI Regulation 2026: Federal Preemption Battles and State Law Patchwork
