Is AI-Powered Tutoring Allowed Under China’s Education Regulations?

Artificial intelligence in education is a rapidly evolving field globally, and China’s regulatory approach to AI-powered tutoring reflects a carefully calibrated balance between encouraging technological innovation and maintaining strict control over education content, data privacy, and social stability. The regulatory landscape is complex and has been evolving rapidly — particularly since the “Double Reduction” policy (2021), the Generative AI Regulations (2023), and subsequent updates.

The short answer: AI-powered tutoring is allowed in China, but it is subject to significant restrictions on what subjects can be taught, what data can be collected, how AI models must be trained, and who can provide AI tutoring services to minors.

Regulatory Framework Overview

AI-powered tutoring in China is governed by at least four overlapping regulatory regimes:

What Is Permitted vs. Restricted

Permitted AI Tutoring Applications

• STEM and coding education: AI-powered platforms teaching programming (Python, Scratch, robotics), mathematics (advanced beyond grade level), science experiments, and engineering concepts are generally permitted. China’s government actively encourages AI in these areas.
• Language learning (English and other foreign languages): AI-powered language tutoring for English and other foreign languages is permitted. Platforms like Liulishuo (流利说), which uses AI for English speaking assessment, operate openly.
• Study aids and homework support for non-academic subjects: AI tools that assist with research, writing practice, and self-directed learning in approved subject areas are allowed.
• Enterprise and professional training: AI tutoring for workplace skills, professional certifications, and continuing education is unrestricted for adult users.
• Assessment and diagnostics: AI systems that assess student learning levels and provide diagnostic feedback are permitted, provided they do not generate subjective evaluations that could be used for competitive ranking.
• Personalized learning pathways: AI platforms that recommend learning materials and create adaptive study plans based on student performance are generally allowed, subject to content review requirements.

Restricted or Prohibited AI Tutoring Applications

• K-9 academic subject tutoring (数学, 语文, 英语 for compulsory education): AI-powered tutoring for core academic subjects aligned with the national curriculum is effectively prohibited under the Double Reduction policy. This includes AI systems that directly teach, quiz on, or provide homework help for compulsory education curriculum content.
• Direct answer generation for homework: AI systems that generate direct answers to homework questions (as opposed to guiding learning) have been repeatedly targeted in regulatory actions. Several “homework AI” apps were removed from Chinese app stores in 2023–2024.
• Content violating socialist core values: Any AI tutoring system that generates content deemed politically sensitive, historically revisionist, or contrary to “socialist core values” (社会主义核心价值观) faces immediate suspension and penalties.
• Unregistered AI services: As of August 2023, all generative AI services accessible to the public in China must be registered with the Cyberspace Administration of China (CAC). Unregistered AI tutoring systems are illegal.
• Excessive data collection: AI tutoring systems collecting biometric data (facial recognition, voiceprints), location tracking, or extensive behavioral data from minors without explicit parental consent face PIPL penalties.

Registration and Compliance Requirements for AI Tutoring Platforms

1. Generative AI Algorithm Registration

Under the Interim Measures for the Administration of Generative AI Services, AI tutoring platforms must:

• Register their AI algorithm with the CAC through the Algorithm Registration System (算法备案系统)
• Submit the algorithm’s training data sources, model architecture, and content filtering mechanisms
• Implement real-time content filtering that blocks prohibited content before users see it
• Maintain logs of AI-generated content for at least 6 months
• Provide users with the ability to flag problematic AI-generated content
• Undergo security assessment if the AI model has “public opinion” or “social mobilization” capabilities (a broad category that includes most interactive educational AI)

The algorithm registration process typically takes 2–4 months and requires technical documentation from the AI development team. As of early 2026, over 2,000 generative AI algorithms had been registered, of which approximately 15% are classified as “education” or “learning” applications.

2. ICP License Requirements

Any AI tutoring platform accessible via the internet must hold an Internet Content Provider (ICP) License (增值电信业务经营许可证). This is the same license required for any commercial website or app operating in China. The application process takes 1–2 months.

3. Education License Considerations

If the AI tutoring platform provides structured, curriculum-aligned instruction (as opposed to being a general study aid), it may need an Online Education License (在线教育办学许可证). This requirement is applied inconsistently across provinces, but the trend is toward stricter enforcement. As of 2026:

• Pure AI tools that provide supplementary learning (e.g., flashcard apps, dictionary tools, grammar checkers) generally do not need an education license
• AI systems that deliver structured courses, assess student progress against curriculum standards, or issue certificates of completion likely need an education license
• AI tutoring platforms targeting K-12 students in academic subjects definitely need an education license — but obtaining one is extremely difficult for for-profit entities

4. Data Protection Requirements

AI tutoring platforms serving Chinese users must:

• Store all user data on servers physically located in mainland China
• Obtain explicit parental consent for users under 14
• Provide data access and deletion mechanisms compliant with PIPL
• Conduct a Personal Information Protection Impact Assessment (PIPIA) before processing data of minors
• Appoint a Data Protection Officer (DPO) with a physical presence in China
• Not transfer student learning data outside of China without passing a security assessment

5. Content Review Mechanisms

AI tutoring platforms must implement multi-layered content review:

• Pre-generation filtering: Block prohibited topics from being queried
• Post-generation review: AI-generated educational content must be scanned for compliance before delivery to the user
• Human review: Platforms serving more than a certain number of users (threshold varies by province) must have dedicated human content reviewers
• Reporting mechanism: Users must be able to report inappropriate content, and platforms must respond within 24 hours

The AI Education Market in China: Key Players

Strategic Considerations for Foreign AI Education Companies

Market Entry Pathways

1. B2B school services (recommended for most foreign entrants): License AI tutoring technology to Chinese schools and education groups rather than operating a direct-to-consumer platform. Schools have existing licenses and can navigate content review more effectively. The government’s “Smart Education” initiative creates strong demand for AI tools in STEM, language learning, and assessment.
2. Enterprise/B2B corporate training: Offer AI-powered professional development and skills training to foreign-invested enterprises in China. This avoids education-specific regulation entirely.
3. Partnership with Chinese AI education companies: Many Chinese education tech companies (iFLYTEK, NetDragon, TAL Education Group) actively seek foreign AI technology partnerships. Licensing your AI model to a Chinese partner with existing regulatory approvals is often faster and safer than establishing a China entity.
4. Non-academic consumer app: Launch AI tutoring for permitted categories — English language learning, coding for kids, music/art education, study skills. These face lighter regulation but must still comply with AI registration, data protection, and content review requirements.

What NOT To Do

• Do NOT launch an AI tutoring service targeting K-9 academic subjects without first obtaining a clear legal opinion from a Chinese education law specialist. The regulatory risk is very high.
• Do NOT use AI models trained on unapproved data sources. All training data for AI used in China must be screened for prohibited content.
• Do NOT assume that an AI system available globally (e.g., Khan Academy’s Khanmigo, Duolingo Max) can be deployed in China without modification. Content filtering, data localization, and algorithm registration are mandatory.
• Do NOT collect student data without a local entity, local servers, and proper consent mechanisms. PIPL penalties can reach RMB 50 million or 5% of annual global revenue.

Recent Enforcement Actions

Case: Homework AI Apps (2023–2024)

In 2023, the CAC and Ministry of Education jointly conducted a sweep of AI homework-help applications. Over 30 apps were removed from Chinese app stores for violating the Double Reduction policy by providing direct answers to K-9 homework questions. Some of these apps used generative AI to produce step-by-step solutions, which was deemed equivalent to academic tutoring. The crackdown extended to apps that disguised homework help through “study companion” interfaces.

Case: Algorithm Registration Enforcement (2024)

In early 2024, the CAC publicly named 17 AI companies that had failed to register their algorithms. Several education AI platforms were among them. Penalties included fines (RMB 100,000–500,000), temporary service suspension, and orders to stop providing services until registration was completed.

Future Outlook

The regulatory direction for AI tutoring in China is toward greater structure, not greater openness. Key trends to watch:

• Subject-specific AI guidelines: The Ministry of Education is reportedly developing subject-by-subject guidelines for AI use in education, with STEM subjects receiving the most permissive treatment and humanities/social sciences the most restrictive.
• Licensing requirement clarification: Rules expected in 2026–2027 will likely clarify exactly which AI tutoring applications require an education license, reducing the current grey-area ambiguity.
• AI grading and assessment restrictions: New regulations may limit AI’s role in high-stakes student assessment, particularly for the gaokao (college entrance exam) and zhongkao (high school entrance exam).
• National AI education platform: China is developing a state-backed AI education platform that will set content standards and potentially compete with — or require integration from — commercial AI tutoring systems.
• Cross-border data transfer: Tightening of data localization enforcement will make it harder for foreign AI tutoring platforms to operate without local infrastructure.

Conclusion

AI-powered tutoring is legal in China but operates within a carefully controlled regulatory environment that reflects the government’s dual objectives: promoting AI innovation in education while maintaining strict control over educational content and data. The most significant restriction is the effective ban on for-profit AI tutoring for K-9 academic subjects under the Double Reduction policy. However, substantial opportunities exist in STEM education, language learning, vocational training, enterprise education, and B2B technology licensing to Chinese schools.

Foreign AI education companies should prioritize the B2B school channel (where demand from China’s “Smart Education” initiatives is strong) and ensure full compliance with AI registration, data protection, and content review requirements before launching any consumer-facing product. Working with a Chinese partner who has existing regulatory relationships is strongly advisable for navigating this complex landscape.