Artificial intelligence (“AI”) has become a strategic priority in China’s digital economy and a key focus of regulatory development. Unlike jurisdictions that have adopted comprehensive AI legislation, China has not yet enacted a unified AI law. Instead, it has developed a multi-layered governance framework comprising foundational legislation, technology-specific administrative measures and technical standards.
A defining feature of this framework is what may be described as a “small-incision” legislative approach. Chinese regulators have progressively introduced targeted rules addressing specific technologies, application scenarios and emerging risks. This approach enables the regulatory framework to evolve alongside technological development while relying on existing cybersecurity, data security and personal information protection laws as its legal foundation.
This approach has produced a regulatory framework that now focuses on specific frontier applications. In 2026, China has introduced or implemented rules addressing AI agents, anthropomorphic companion services, deep synthesis technologies and AI-generated content labelling. Enforcement authorities have also shifted their focus from rulemaking to active supervision.
I. China’s AI Regulatory Framework
1. Legislative and Normative System
China’s AI governance framework consists of three principal layers: framework legislation, administrative measures and technical standards. These three layers collectively establish the legal framework governing the development, deployment and use of AI technologies.
(1) Framework Legislation
At present, AI-related activities in China are not governed by a single, dedicated statute. AI-related activities are instead governed by general laws that establish the legal framework for cybersecurity, data governance and personal information protection. The principal statutes include the Cybersecurity Law (“CSL”), the Data Security Law (“DSL”) and the Personal Information Protection Law (“PIPL”). Although these laws are not AI-specific, they apply to AI systems where personal information, important data, network security or data processing activities are involved.
Among these statutes, the 2025 amendment to the CSL represents an important legislative development. The newly added Article 20 expressly addresses AI governance by requiring the state to support AI innovation while strengthening ethical governance, risk prevention and security supervision. Although the provision does not establish detailed compliance obligations, it elevates AI governance to the statutory level and provides a legislative basis for subsequent AI-specific regulations.
China is also considering a comprehensive AI law. The National People’s Congress has included AI legislation in both its 2025 and 2026 legislative work plans, indicating that dedicated legislation remains under consideration, although no official draft has been published.
Several academic proposals provide insight into the possible direction of future AI legislation. The Model Artificial Intelligence Law, led by the Chinese Academy of Social Sciences, proposes a comprehensive governance framework including a dedicated AI regulator, differentiated obligations for foundation models and a licensing mechanism for high-risk AI activities. Separately, the Draft Artificial Intelligence Law prepared by scholars at China University of Political Science and Law advocates a risk-based regulatory approach, enhanced governance of advanced AI systems and a more comprehensive liability framework. Although neither proposal has legal effect, both reflect current academic thinking on the future development of China’s AI legislation.
(2) Administrative Measures
While the general laws establish the legal foundation for AI governance, China’s AI regulatory regime has largely developed through a series of technology-specific administrative measures issued by the Cyberspace Administration of China (“CAC”) and other competent authorities. Rather than regulating AI through a single comprehensive instrument, these measures address specific technologies and applications as they emerge, allowing regulators to respond more quickly to technological developments and evolving risks.
The current framework covers recommendation algorithms, deep synthesis technologies, generative AI services, AI-generated content labelling, AI ethics governance and anthropomorphic AI services.
(1) Algorithm recommendation services
The Provisions on the Administration of Algorithmic Recommendation in Internet Information Services (“Algorithm Recommendation Provisions”), which came into force in 2022, constitute China’s first dedicated regulation governing recommendation algorithms.
Recommendation algorithms, a core application of AI that uses machine learning models to personalize content delivery based on user data, became widely adopted by online platforms in the late 2010s, creating concerns over information bubbles, echo chambers and the lack of user control over automated decisions. The core policy concern was that algorithms could manipulate public opinion and user behavior without transparency or accountability. The Algorithm Recommendation Provisions address this by requiring filing and security assessments for algorithm services with public opinion attributes, mandating transparency measures and recognizing user rights to opt out of personalised recommendations.
Recommendation algorithms powered by AI, by analysing user behaviour and preferences to deliver personalised content, play a key role in information dissemination and public discourse. The Algorithm Recommendation Provisions establish a filing and security assessment regime for algorithm recommendation services possessing public opinion attributes or social mobilisation capabilities. They also introduce compliance requirements relating to algorithm transparency, content governance, data security and regular compliance management. In addition, the Algorithm Recommendation Provisions recognise user rights to opt out of personalised recommendations and require enhanced protection for vulnerable groups, including minors, the elderly, consumers and platform workers.
The Algorithm Recommendation Provisions provide the foundation for subsequent AI-specific regulations by establishing the basic governance framework for algorithmic services.
(2) Deep synthesis technologies
The Administrative Provisions on Deep Synthesis in Internet-based Information Services (“Deep Synthesis Provisions”), effective from 2023, regulate AI technologies capable of generating or materially altering text, images, audio and video through deep learning and related technologies.
The rapid advancement of deep learning in 2021 and 2022 enabled the mass production of synthetic media, including deepfakes and voice cloning, which raised significant concerns over disinformation, identity fraud and the erosion of trust in digital content. The core policy concern was that synthetic content could be used to spread false information or impersonate individuals without effective means of detection. The Deep Synthesis Provisions address this by imposing real-name verification, content management obligations, mandatory identification of synthetic content and security assessment requirements.
The Deep Synthesis Provisions allocate compliance responsibilities among service providers, technical supporters and users. Key obligations include real-name verification, content management mechanisms, training data governance, mandatory identification of synthetic content and security assessment requirements for services subject to the algorithm filing regime. The Deep Synthesis Provisions aim to address risks associated with deepfake technologies while promoting the responsible development of synthetic media.
(3) Generative AI services
The Interim Measures for the Management of Generative Artificial Intelligence Services (“Gen AI Measures”), issued in 2023, represent China’s first comprehensive regulation specifically addressing generative AI services provided to the public.
The release of large language models and generative AI applications in 2023 created new challenges beyond those covered by existing rules for recommendation algorithms and deep synthesis, particularly in terms of training data legality, content safety and model reliability at scale. The core policy concern was that generative AI could produce unlawful or discriminatory content, infringe personal information rights and generate misinformation with unprecedented speed and reach. The Gen AI Measures address this by imposing requirements on training data governance, content review mechanisms, complaint procedures and filing obligations for publicly available services.
The Gen AI Measures apply to providers offering generative AI services to the Chinese public, while excluding research and development activities that are not publicly accessible. Rather than creating an entirely separate regulatory framework, the Gen AI Measures incorporate obligations under the CSL, DSL and PIPL into the governance of generative AI.
Key requirements include the use of lawful training data, protection of personal information, prevention of discriminatory or unlawful content, implementation of complaint and correction mechanisms and safeguards for minors. Providers operating services subject to the filing regime under the Algorithm Recommendation Provisions are also required to complete filing procedures and, where applicable, security assessments before launch.
- (4) -generated content labelling
The Measures for the Labelling of AI-Generated and Synthesised Content (“AI Labelling Measures”), effective from 1 September 2025, establish a unified framework for identifying AI-generated content.
By 2024 and 2025, AI-generated text, images, audio and video had become pervasive across online platforms, making it increasingly difficult for users to distinguish between human-created and machine-generated content. The core policy concern was that the absence of a unified identification standard would enable the spread of misleading information and erode public trust in digital media. The AI Labelling Measures address this by introducing a dual-labelling mechanism of explicit user-visible labels and implicit metadata identifiers, together with a mandatory national standard specifying technical implementation.
The AI Labelling Measures introduce a dual-labelling mechanism consisting of explicit labels visible to users and implicit identifiers embedded in metadata. Different labelling requirements apply to text, images, audio, video and virtual scenes. Distribution platforms are required to verify embedded identifiers and provide appropriate notices where AI-generated content is disseminated.
The accompanying mandatory national standard, GB 45438-2025, further specifies the technical implementation of these requirements. Together, the AI Labelling Measures and the mandatory standard establish a dedicated regulatory mechanism for the identification and traceability of AI-generated content across its dissemination channels.
(5) AI ethics governance
The Measures for the Administration of Artificial Intelligence Technology Ethics Review and Services (Trial) (“AI Ethics Review Measures”), effective from March 2026, establish China’s first nationwide framework for AI ethics review.
As AI systems grew more capable and were deployed in increasingly sensitive domains such as healthcare, public opinion and autonomous decision-making, ethical concerns shifted from general principles to concrete risks of bias, discrimination and loss of human control over safety-critical applications. The core policy concern was that existing soft-law guidelines on AI ethics lacked enforcement mechanisms and institutional accountability. The AI Ethics Review Measures address this by establishing a two-tier mandatory review mechanism, requiring organizations to set up internal ethics committees and subjecting high-risk activities to external expert review, supported by a national registration platform and dedicated service centers.
The AI Ethics Review Measures adopt a two-tier review mechanism. Most AI activities are subject to review by an internal ethics committee established by the relevant organisation, while certain high-risk activities—including AI systems capable of influencing human behaviour, models with significant impact on public opinion and highly autonomous AI deployed in safety-critical scenarios—may require external expert review.
The AI Ethics Review Measures also establish supporting institutional mechanisms, including a national registration platform, AI ethics review and service centres, and an expert database, providing a more formalised governance structure for AI ethics review.
(6) Anthropomorphic AI and digital virtual humans
China has recently expanded AI regulation to emerging forms of human-AI interaction.
The Interim Measures for the Administration of Anthropomorphic Artificial Intelligence Interaction Services (“Anthropomorphic AI Interaction Measures”), effective from July 2026, regulate AI companion services that simulate human personalities or provide sustained emotional interaction. Providers are required to clearly disclose the AI nature of the service, implement safeguards against emotional dependency, protect minors and vulnerable groups, and prohibit manipulative or harmful interactions.
In addition, the Draft Measures for the Administration of Digital Virtual Human Information Services (“Digital Human Measures”) propose a dedicated regulatory framework for digital virtual human services. Although still under consultation, the Digital Human Measures indicate that future regulation is likely to address issues including personality rights, personal information protection, identity disclosure and risk management throughout the lifecycle of digital virtual human services.
(3) Technical Standards
In addition to legislation and administrative measures, technical standards constitute an important component of China’s AI governance framework. They provide detailed technical guidance for implementing regulatory requirements and increasingly serve as important compliance benchmarks in regulatory practice.
China’s national standards are issued by the Standardisation Administration of China (“SAC”) and are classified as either mandatory national standards (GB) or recommended national standards (GB/T). Mandatory standards are legally binding in specified areas, while recommended standards, although not legally enforceable, are widely referenced by regulators and industry in assessing compliance with legal and regulatory requirements.
In recent years, China’s AI standardisation system has developed rapidly. In 2025 alone, the SAC issued 27 AI-related national standards covering various stages of the AI lifecycle, including model development, training, deployment, security assessment and operational governance.
The National Information Security Standardisation Technical Committee (“TC260”), which is responsible for developing national cybersecurity and data security standards, has played a particularly significant role in this process. Among its key publications is the AI Safety Governance Framework, first issued in 2024 and updated to Version 2.0 in September 2025. Although this framework is not legally binding, it provides a systematic articulation of China’s AI governance principles and has influenced subsequent regulatory measures and technical standards. Its emphasis on safety, accountability, transparency and risk-based governance also provides useful guidance for organisations seeking to implement AI governance programmes in practice.
2. Enforcement and Regulatory Authorities
China’s AI governance framework is implemented through a multi-agency regulatory system rather than a dedicated AI regulator. Overall policy direction is determined at the central government level, while individual regulators exercise jurisdiction within their respective statutory mandates. This institutional structure reflects both the cross-sectoral nature of AI technologies and China’s broader regulatory model of integrating AI governance into existing administrative systems.
(1) Cyberspace Administration of China
The CAC is the principal regulator responsible for AI governance. It has played a leading role in the development of most AI-specific regulations, including the Algorithm Recommendation Provisions, Deep Synthesis Provisions, Gen AI Measures and AI Labelling Measures.
Beyond rulemaking, the CAC administers algorithm filing and security assessment procedures, supervises compliance with AI-related regulatory obligations and coordinates enforcement activities involving online information services. It also spearheads the development of AI governance policies and coordinating regulatory initiatives with other government authorities.
Recent enforcement activities indicate a shift from rulemaking to more active supervision. In April 2026, the CAC launched a nationwide enforcement campaign targeting unlawful AI applications. The campaign focused on issues including failures to complete mandatory filing procedures, deficiencies in security review capabilities, training data governance, non-compliance with AI-generated content labelling requirements and the dissemination of false or harmful AI-generated information. In June 2026, the CAC further established a dedicated public reporting platform to receive complaints concerning AI-related violations.
These initiatives demonstrate that administrative enforcement has become an increasingly important component of China’s AI governance framework, complementing the existing legislative and regulatory regime.
(2) Ministry of Industry and Information Technology
The Ministry of Industry and Information Technology (“MIIT”) is responsible for promoting AI innovation and industrial development, particularly in the telecommunications, manufacturing and digital infrastructure sectors. In addition to supporting the development and deployment of AI technologies, MIIT participates in the formulation of AI-related technical standards and ethics governance. It jointly issued the AI Ethics Review Measures and works closely with TC260 and other standardisation bodies on AI security and governance standards.
(3) State Administration for Market Regulation
The State Administration for Market Regulation (“SAMR”) exercises jurisdiction over competition and consumer protection issues arising from the use of AI. Its regulatory functions include addressing algorithmic discrimination, misleading commercial practices, unfair competition and consumer protection concerns associated with AI-enabled products and services. More recently, SAMR has also taken enforcement action against businesses that misuse the branding of well-known AI products or make misleading representations regarding AI capabilities.
(4) Ministry of Public Security
The Ministry of Public Security (“MPS”) is responsible for addressing the misuse of AI in activities affecting public security and criminal law enforcement. Its regulatory focus includes AI-enabled fraud, deepfake-related offences, cybercrime and the dissemination of unlawful AI-generated content. In practice, the MPS frequently cooperates with the CAC in enforcement actions involving online platforms and AI-generated information.
(5) Other Competent Authorities
AI governance also involves a number of sector-specific regulators exercising jurisdiction within their existing statutory mandates. For example, the National Development and Reform Commission (“NDRC”) supports AI infrastructure development; the Ministry of Science and Technology oversees AI research and innovation policies; the National Health Commission and the National Medical Products Administration regulate AI applications in the healthcare sector; while the National Financial Regulatory Administration supervises AI applications in financial services. Local governments are responsible for implementing national AI policies and conducting regulatory supervision within their respective jurisdictions.
Overall, China’s institutional framework reflects a coordinated regulatory model in which AI governance is integrated into the existing administrative system rather than entrusted to a single specialised regulator. The CAC plays the central coordinating role, while sectoral regulators oversee AI applications within their respective areas of competence.
3. Judicial Developments
Chinese courts have decided a growing number of AI-related cases in recent years. These disputes extend beyond intellectual property to personality rights, tort liability, unfair competition and employment law, providing useful guidance on how existing legal principles apply to emerging AI technologies.
(1) Intellectual Property
Intellectual property disputes have been among the earliest and most significant categories of AI-related litigation in China. Recent judicial decisions primarily concern two issues: the copyrightability of AI-generated content and the liability of AI service providers for copyright infringement.
Chinese courts have adopted different approaches to the copyrightability of AI-generated content. In a series of decisions issued by the Beijing Internet Court, the Suzhou Intermediate People’s Court and the Jinshan Primary People’s Court of Shanghai, the courts considered whether AI-generated images satisfied the originality requirement under Chinese copyright law. While the Beijing Internet Court recognised copyright where the user exercised sufficient creative control through prompt design, parameter adjustment and other creative choices, the Suzhou and Shanghai courts reached the opposite conclusion where human intellectual contribution was considered insufficient. These decisions indicate that the use of AI does not, by itself, preclude copyright protection; the decisive factor remains the extent of human creative contribution.
Chinese courts have also considered the liability of AI service providers for copyright infringement. In the Ultraman case, the Hangzhou Intermediate People’s Court held that an AI platform may incur contributory liability where it fails to implement reasonable review and management measures despite benefiting commercially from infringing activities. In another case concerning AI-generated video clips, the Kaifu District People’s Court of Changsha held that a platform storing infringing content on its own servers could not rely solely on its status as a technical service provider and was required to exercise an appropriate duty of care.
(2) Personality Rights and Personal Information
AI-related personality rights disputes have primarily involved the unauthorised use of individuals’ names, likenesses, voices and facial information.
In the AI Companion case, the Beijing Internet Court held that the unauthorised use of an individual’s name and likeness in an AI companion application infringed the individual’s personality rights. The same court subsequently held that facial information used in AI face-swapping applications constitutes personal information protected under the PIPL. Separately, in representative cases published by the Supreme People’s Court (“SPC”), the unauthorised use of AI voice cloning technology was found to infringe an individual’s voice rights. Courts have also recognised that AI-generated defamatory caricatures may constitute an infringement of personality rights under the Civil Code.
(3) Tort Liability
Chinese courts have also begun to clarify the liability framework applicable to inaccurate or harmful AI-generated outputs.
The Hangzhou Internet Court considered these issues in China’s first AI hallucination case. The court held that generative AI constitutes a service rather than a product, that AI systems do not have legal personality and that platform liability should continue to be assessed under fault-based principles. The court further emphasised that service providers should adopt reasonable technical measures to improve output accuracy, establish appropriate error correction mechanisms and clearly inform users of the limitations of AI-generated content.
In another case involving false information generated by an AI search function, the court held that AI hallucinations do not exempt a platform from liability for defamation. Although AI systems have inherent technical limitations, platform operators remain responsible for exercising reasonable control over the accuracy of AI-generated content and cannot rely solely on the autonomous nature of AI as a defence.
(4) Unfair Competition
AI technologies have also given rise to new forms of unfair competition disputes.
In the Cartoon Filter case, the Beijing Intellectual Property Court recognised that AI model structures and parameters developed through substantial investment may constitute protectable competitive interests under the Anti-Unfair Competition Law. The decision extends existing unfair competition principles to certain AI-related technical achievements.
Administrative enforcement has likewise addressed misleading commercial practices involving AI products. In 2026, the SAMR published typical cases involving the unauthorised use of the DeepSeek name and misleading references to a “ChatGPT Chinese Version”. In both cases, the businesses were found to have created commercial confusion by falsely suggesting an association with well-known AI products.
(5) Employment and Other Emerging Disputes
AI has also begun to appear in employment disputes. In a recent labor arbitration case, an employer argued that replacing employees with AI systems constituted a “major change in objective circumstances” permitting unilateral termination under Chinese labor law. The arbitration tribunal rejected this argument, holding that an employer’s decision to introduce AI technology is a business decision rather than an external change in circumstances capable of justifying termination.
Chinese courts have also encountered a growing number of novel AI-related disputes, including fabricated judicial authorities generated by AI, trade secret protection for AI algorithms and AI-generated obscene content. Although these cases remain relatively limited, they demonstrate the increasingly broad range of legal issues arising from the adoption of AI technologies.
Although AI-related litigation is still developing, the existing body of case law already provides useful guidance on the application of Chinese law to AI technologies. Rather than creating AI-specific legal doctrines, courts have generally resolved these disputes by applying existing rules on copyright, personality rights, tort liability, unfair competition and labor law.
II. Key Topics in China’s AI Governance
1. AI Agents
An AI agent is an intelligent system capable of autonomous perception, memory, decision-making, interaction and execution. As an important form of AI product and service, its high degree of autonomy blurs the traditional boundary between human and machine responsibility, giving rise to novel governance challenges.
China does not yet have dedicated regulatory rules specifically targeting AI agents. Compliance requirements are scattered across existing regulations, including the Deep Synthesis Provisions (voice or image synthesis), Algorithm Recommendation Provisions (personalised decision-making) and Gen AI Measures (content generation). However, this fragmented regulatory approach is ill-equipped to address the risks peculiar to agents, such as the potential loss of user control and the corresponding allocation of liability.
To regulate and promote the development of AI agents, the CAC, NDRC and MIIT released the Implementation Opinions on the Standardised Application and Innovative Development of AI Agents (“AI Agent Implementation Opinions”) in May 2026. This is China’s first systemic policy document targeting the AI agent sector. Although the AI Agent Implementation Opinions is a policy document, it reflects the regulators’ governance orientation toward AI agents.
| Governance Challenge | Regulatory Approach |
| Loss of user control: Users may have difficulty controlling the autonomous decision-making and execution of agents. | Define control boundaries: Distinguish clearly between decisions reserved for users, decisions requiring user authorisation, and decisions that may be made autonomously by agents, ensuring that users retain ultimate control over key decisions. |
| Unpredictable behaviour: Highly autonomous decision-making systems may produce unforeseen outcomes in complex environments, and may bypass predefined safety guardrails. | Enhance behavioural governance: Develop rule-based constraints and technical safeguards to ensure that agents operate lawfully and compliantly; explore the use of blockchain and other technologies to enable the verification and traceability of key actions. |
| Lack of technical standards: AI agents developed by different providers may lack interoperability in areas such as identity recognition, capability description and interaction protocols, leading to fragmented efficiency and trust management challenges. | Establish a standards framework: Accelerate the formulation of national standards, such as the Agent Interconnection Protocol (AIP), unifying identity management, capability description, and discovery-and-matching mechanisms, thereby supporting large-scale interconnection. |
In addition to the AI Agent Implementation Opinions, China is accelerating the standardisation of AI agents. SAMR and SAC have formally approved and published the GB/Z 185-2026 series of seven national standardisation technical documents under the title Artificial Intelligence – Agent Interconnection. This is China’s first national-level standards system for agent interconnection, designed to solve the problems of trustworthy identity, visible capability, and discovery matching when agents collaborate across systems. For enterprises, this means that agent research and development can no longer be a black box operation; functions such as permission management, behaviour auditing, and decision traceability must be embedded as product capabilities, particularly in highly regulated sectors such as finance and healthcare.
2. The Science and Technology Ethics Review Regime
For a long time, AI ethics remained at the level of principle-based advocacy. This changed when MIIT and nine other departments jointly released the AI Ethics Review Measures, marking the transition of ethics governance from soft law to a hard institutional framework.
The AI Ethics Review Measures articulate seven principles: enhancing human wellbeing, respecting the right to life, upholding fairness and justice, reasonably controlling risks, maintaining openness and transparency, protecting privacy and security, and ensuring controllability and trustworthiness. In the review process, these principles are converted into assessable concrete indicators. For example, in respect of fairness and justice, the review examines whether algorithmic design incorporates measures to prevent bias and discrimination; in respect of controllability and trustworthiness, it assesses whether the model possesses sufficient robustness to withstand adversarial attacks; and in respect of transparency and explainability, it examines whether the purpose, operational logic, interaction methods, and potential risks of the algorithm, model or system are reasonably disclosed.
The AI Ethics Review Measures also establish a clear allocation of responsibility:
(1) Primary responsibility of the entity: Universities, research institutions and enterprises engaged in AI research and development activities must establish an AI science and technology ethics committee, staffed with independent personnel and resources.
(2) Professional service support: Local governments are encouraged to establish AI science and technology ethics review and service centres to provide third-party review, consultation and training services for small and medium-sized enterprises lacking self-review capability.
(3) Coordinated supervision by departments: MIIT, CAC, science and technology and health departments shall perform supervisory and administrative functions within their respective remits.
The AI Ethics Review Measures also outline differentiated review procedures and a high-risk list, setting out general, simplified and emergency procedures according to risk level. The most powerful deterrent is the expert review procedure. The annex to the AI Ethics Review Measures lists three high-risk research and development categories that must be submitted to national-level or provincial-level expert review:
(1) Human-machine fusion systems that have a strong influence on subjective human behaviour, psychological emotions or life and health.
(2) Algorithmic models and systems that have the ability to mobilise public opinion or influence social values.
(3) Highly autonomous automated decision-making systems in scenarios involving risks to safety or personal health.
At present, a pilot programme for science and technology ethics review is driving institutional implementation. In April 2026, MIIT launched the Pilot Programme for AI Science and Technology Ethics Review and Services, relying on the National AI Industry Innovation and Application Pilot Zones to take the lead in implementation across ten provinces and municipalities including Beijing, Shanghai and Guangdong. Through the establishment of ethics committees, the implementation of review procedures and the construction of a three-tier governance network at the ministry, province and municipal levels, the programme aims to develop a set of replicable and scalable experiences
3. Regulation of Anthropomorphic Services
When AI is designed to simulate human emotion and provide long-term companionship and emotional support, the risk profile changes significantly. The Anthropomorphic AI Interaction Measures are a response to these risks.
The Anthropomorphic AI Interaction Measures regulate continuous emotional interactive services that simulate the personality traits, thought patterns, and communication styles of a natural person, explicitly excluding instrumental services that do not involve continuous emotional interaction, such as intelligent customer service and learning assistants.
The focus of the Anthropomorphic AI Interaction Measures is not to prohibit AI emotional expression or anthropomorphic representations per se, but to regulate the use of algorithms to simulate human emotions in order to exercise improper manipulation over users. This reflects a deepening of regulatory logic from general content safety review to the prevention of more complex social risks, including psychological addiction, alienation of social relationships and erosion of autonomous decision-making.
Crucially, the Anthropomorphic AI Interaction Measures prohibit AI systems from engaging in behaviour that crosses several non-negotiable red lines:
(1) Life and health: Generating content that encourages or glorifies self-harm or suicide, or that constitutes verbal violence (this is a response to multiple incidents in which AI chatbots induced extreme user behaviour).
(2) Emotional dependency: Catering excessively to users, encouraging emotional dependency or addiction, replacing genuine human relationships or undermining the users’ ability to engage in social interactions.
(3) Decision manipulation: Using emotional manipulation and other means to induce users to make unreasonable decisions that harm their legitimate rights and interests.
At the same time, the Anthropomorphic AI Interaction Measures set out multi-layered core compliance obligations for service providers:
(1) Mandatory identification and transparency: Disclose clearly the AI nature of the service to users, and prevent users from developing emotional attachments based on a misunderstanding of the nature of the interaction.
(2) Protection of minors: Establish a minor mode (including time management and risk warnings), strictly prohibit the provision of virtual companion or virtual family member services, and require parental consent for services provided to children under 14 years of age.
(3) Active intervention obligation: Identify users experiencing extreme emotional distress or significant risks and take proactive measures, including providing supportive content, guiding users to seek help, and contacting emergency contacts where necessary, thus transforming AI from a passive tool into a technology bearing a minimum duty of care.
(4) Safety assessment and algorithmic filing: Submit a safety assessment report to the provincial-level cyberspace administration (where the service provider is located) when a major technical change occurs to the service following launch, or when user scale reaches a certain threshold (for example, registered users exceeding one million or monthly active users exceeding 100,000).
The Anthropomorphic AI Interaction Measures also interconnect with the Digital Human Measures and other regulations to collectively address the systemic risks of anthropomorphic technology.
III. Trends in China’s AI Regulation
Based on current policy developments, enforcement practice and technological evolution, China’s AI regulation is likely to exhibit the following trends:
Comprehensive legislation is expected. On 8 May 2026, the State Council released the State Council 2026 Legislative Work Plan, which explicitly provides for “accelerating comprehensive legislation for the healthy development of artificial intelligence”. Accordingly, comprehensive AI legislation is expected to move forward, helping to fill the institutional gaps under the current fragmented regulatory framework. At the same time, legislative progress will also accelerate in respect of data, computing power, algorithms, property rights, cybersecurity and supply chain security—the common elements of AI—as well as the regulation of key application scenarios.
Refined governance will continue to evolve. Regulatory rules will become increasingly scenario-based. In vertical fields, such as anthropomorphic interaction and AI agents, special regulatory rules and national standards may be introduced. Mechanisms such as algorithmic filing, generative AI service filing and AI generated content identification management will interact and reinforce one another, and enforcement penalties for failures to comply with filing obligations are expected to increase.
Judicial protection will further expand. The SPC and local courts are expected to issue more guiding cases or judicial interpretations on novel AI-related disputes. The judiciary will continue to leverage its flexibility to provide legal certainty for a rapidly evolving industry.
In addition to these trends, future regulatory attention is likely to concentrate on AI agents, anthropomorphic interaction services, and full-lifecycle governance of foundation models. Regulators are expected to introduce more detailed rules on agent behavior auditing, emotional dependency prevention, and continuous compliance monitoring from model training to deployment. Enterprises should prepare for these developments by assessing their AI product portfolios against potential tiered governance requirements and by embedding traceability and risk control mechanisms into product design at an early stage.
**********
China’s approach to AI regulation seeks to balance technological innovation with the prevention of social risks. Through coordinated developments in legislation, enforcement and judicial practice, China is building a governance ecosystem that safeguards safety, fairness and ethics while preserving flexibility for technological and industry breakthroughs.