China’s AI Regulation in 2026: Enforcement, Guidance and Compliance

New rules are only part of China’s AI governance picture. How regulators enforce those rules, courts interpret them, and sectors apply them determines the compliance risks businesses face in practice. In 2026, all three of these areas saw significant developments.

This is the second article in our two-part series on China’s AI regulatory developments in 2026. In part one, we covered new laws and rules governing AI services, including the planned comprehensive AI law, rules for anthropomorphic AI services, and AI agent governance. In part two, we examine the enforcement of AI content labelling requirements, the SPC’s opinions on AI-related disputes, and the NFRA’s guidance for the banking and insurance sectors.

See part one here: China’s AI Regulation in 2026: New Laws and Rules for AI Services [link to be added]

Enforcement of AI content labelling requirements

Labelling obligations

The Measures for Labelling AI-Generated and Synthetic Content, jointly issued by the Cyberspace Administration of China (CAC), the Ministry of Industry and Information Technology, the Ministry of Public Security, and the National Radio and Television Administration, took effect on 1 September 2025 together with the mandatory national standard GB 45438-2025. The measures require two types of labels:

• Explicit labels, which can be seen or heard by users, must be added to AI-generated text, images, audio, video, and virtual scenes and must be retained when the content is downloaded or exported; and
• Implicit labels must be embedded in the metadata of AI-generated files, with digital watermarks encouraged.

The obligations extend across the content chain. Content distribution platforms must check for implicit labels and add prominent warnings to content that is AI-generated or suspected to be AI-generated. App distribution platforms must verify whether apps offer AI generation services and review their labelling materials. Service providers must explain their labelling methods in their user agreements and retain relevant logs for at least six months where users request content without explicit labels. Users who publish AI-generated content must declare it as such, and the malicious removal, alteration, or forgery of labels is prohibited.

Enforcement in 2026

In 2026, the authorities moved from implementation to active enforcement. In April 2026, the CAC directed local cyberspace authorities to take action against the Jianying and Maoxiang apps and the Jimeng AI website for failing to properly implement labelling requirements. The legal bases cited included the Cybersecurity Law, the Interim Measures for the Management of Generative AI Services, and the labelling measures. The measures taken included summoning company representatives, issuing rectification orders and warnings, and strictly disciplining responsible personnel. The CAC also indicated that it would continue to strengthen oversight of AI content labelling.

The decision to target widely used consumer products indicates that labelling compliance is now a routine supervisory focus. It should also be noted that labelling obligations are not limited to AI developers. Companies that publish AI-generated marketing materials, product images, or videos on Chinese platforms must declare them and ensure labels are not removed during editing or publishing.

Judicial guidance on AI-related disputes

On 7 September 2026, the SPC issued the Opinions on Lawfully Hearing Cases Involving Artificial Intelligence Disputes (Fa Fa [2026] No. 10). The 24 articles are divided into five parts covering general requirements, tort liability, intellectual property, procedural and criminal matters, and working mechanisms, and provide lower courts with a unified approach to AI-related cases.

Liability for AI-generated content

Unless the law provides otherwise, courts will apply the general fault-based liability principle under Article 1165 of the Civil Code to determine liability for AI-related harm. This means that AI service providers are not automatically liable for all content their systems generate.

However, the opinions establish a notice-and-takedown mechanism for generative AI. Where a generative AI service provider fails to promptly take necessary measures, such as stopping the generation of infringing content, after receiving a valid notice from a rights holder, it will be liable for the resulting damage. A valid notice must include preliminary evidence of infringement and the rights holder’s true identity. Users who deliberately prompt infringing content may also be held liable, and courts may issue injunctions requiring generative AI providers to stop the relevant services where personality rights are at risk.

Personal information, voice, and likeness

The opinions state that processing personal information made public for AI model training, within a reasonable scope, generally is not considered an infringement unless the individual has expressly refused. However, consent is required when the processing significantly affects the individual’s rights and interests.

The opinions also address several AI-specific personality rights issues:

• Using a natural person’s voice as training data to generate an identifiable synthetic voice without consent infringes that person’s voice rights;
• The unauthorised use of a person’s name, likeness, or voice in AI face-swapping, voice cloning, and the “digital resurrection” of deceased persons is addressed as a personality rights infringement; and
• Using AI to collect and disclose private information to expose individuals online constitutes a privacy infringement.

Intellectual property and training data

In copyright infringement disputes, courts will determine liability for developers, service providers, and users based on factors such as the sources of training data, their level of participation, the safeguards in place, and the profits obtained. Developers claiming non-infringement must provide information on training data sources, training records, and how the model operates. Users who knew or should have known of an earlier work and use AI to generate a substantially similar work without a reasonable defence may be held liable.

On patents, where a natural person uses AI to complete an invention, that person will be recognised as the inventor if they have made a creative contribution to the substantive features of the invention. Open-source developers may be exempted from liability for shared code modules if they publicly explain the functions and safety risks of those modules.

Data rights and competition

Courts will protect the data rights of AI developers who lawfully obtain data through collection, derivation, transfer, or licensing, with protection available under copyright law, trade secret rules, or the Anti-Unfair Competition Law, depending on the data’s nature. The opinions also address liability for sabotaging AI systems through falsified data, malicious labelling, or adversarial attacks, as well as monopoly agreements and abuse of dominance carried out through data or algorithms.

In consumer-facing cases, courts will assess whether algorithm-driven differential pricing is justified, while AI-enabled fake sales, fake traffic, and false promotion may constitute unfair competition. AI-generated celebrity impersonations used to promote products may trigger punitive damages under the Consumer Rights Protection Law if they amount to fraud.

Evidence and procedure

When AI-generated output is used as evidence of infringement, courts will consider factors including prompt design, the output’s similarity to the protected work, the consistency of results in repeated tests, and the filtering mechanisms in place. In autonomous-driving disputes, courts may order vehicle manufacturers, sellers, and operators to provide the driving data needed to establish the facts of an accident.

Taken together, the opinions confirm that businesses remain accountable for both the data used to train and operate AI systems and the content those systems generate. In practice, a dispute’s outcome will likely depend on the documentation a company can produce, making records of training data sources, complaint handling, and content-filtering mechanisms particularly important.

AI governance requirements for banks and insurers

On 18 June 2026, the NFRA released the Guiding Opinions on the Safe Development and Application of Artificial Intelligence in the Banking and Insurance Sectors. The guidance applies to banking and insurance institutions that develop or apply AI and covers seven areas: governance structure, development and application, data governance, computing power, risk management, capability building, and supervision.

Governance and development

Institutions are expected to strengthen top-level design and unified management of AI, and to establish a full lifecycle management system covering application scenarios and business processes. They should establish development and testing systems, manage end-to-end model deployment, and prudently explore developing financial AI agents.

Data and computing power

Institutions should build data management and operations systems and develop high-quality datasets for specific business scenarios. The guidance also calls for secure, controllable computing infrastructure, and encourages larger institutions to provide computing services to smaller ones.

Risk management

The key risk management requirements include:

• Incorporating AI risks into enterprise-wide risk management, with risk classification and admission controls for high-risk applications;
• Ensuring human oversight and intervention at key points in high-risk use cases;
• Managing outsourcing and supply chain risks arising from third-party models and service providers;
• Improving the robustness, transparency, and explainability of AI models; and
• Addressing cybersecurity, data security, personal information protection, operational resilience, and business continuity.

The NFRA and its local branches will supervise compliance, penalise violations, and periodically assess the effectiveness of the policy. Foreign-invested banks and insurers, as well as vendors supplying AI tools to Chinese financial institutions, should expect these requirements to feature in supervisory reviews and procurement due diligence.

Considerations for companies managing AI compliance risks

As enforcement, litigation, and sector supervision of AI intensify, companies should ensure that their compliance measures can be demonstrated in practice, not only on paper.

Reviewing content labelling workflows

Companies should confirm that AI-generated content published in China carries the required explicit and implicit labels and that editing and publishing tools do not remove them. Companies providing AI generation services should also review their user agreements and log retention practices.

Documenting training data and outputs

In light of the SPC’s opinions, companies should maintain records of training data sources, consents obtained, and content filtering mechanisms, as these are likely to be decisive in the event of a dispute.

Establishing notice-and-takedown procedures

Generative AI service providers should establish procedures to receive, assess, and promptly respond to infringement notices, as failing to act after receiving a valid notice may create liability.

Managing third-party AI vendors

Contracts with AI vendors should require compliance with Chinese labelling, filing, and data protection requirements and allocate liability clearly. This is particularly important for companies in the financial services sector, given the NFRA’s focus on outsourcing and supply chain risks.

How Dezan Shira & Associates can help

Dezan Shira & Associates supports foreign-invested enterprises in managing compliance risks under China’s AI regulatory framework, including AI content labelling requirements, personal information protection obligations, and sector-specific requirements for financial institutions. Our team can help review AI content workflows, data processing and vendor arrangements, internal documentation, and design governance frameworks to meet Chinese regulatory expectations. Contact us to discuss your company’s needs.