Data compliance for foreign companies

The rapid advancement of artificial intelligence (AI) has transformed business landscapes globally, and China is no exception. Foreign companies operating in China are increasingly integrating AI into their business models, driven by the potential to enhance efficiency and innovation. However, this integration comes with significant responsibilities, writes Panpan Tang, particularly regarding data security and privacy protection.

---

In recent years, China has strengthened its regulatory framework for AI, introducing a series of laws and regulations designed to safeguard data and ensure ethical AI use. For foreign companies, navigating these regulations is crucial to avoid legal risks and protect their reputation.

Core legal basis and regulatory framework

To operate in a compliant manner, foreign companies must understand and adhere to China’s key legal and regulatory frameworks governing AI and data.

Basic laws

China’s legal landscape for data and AI is built on three foundational laws:

• The Cybersecurity Law, which focusses on safeguarding network data and cybersecurity. It requires companies to implement technical measures to prevent cyber attacks and data breaches.

• The Data Security Law, which mandates the classification and management of data based on its sensitivity and importance.

• The Personal Information Protection Law, which governs the handling of personal information, emphasising the principles of legality, propriety, necessity and good faith. For cross-border transfers of personal information, it clarifies three compliance pathways: mandatory security assessments (for high-risk transfers), filing standard contracts with competent regulatory authorities, or obtaining personal information protection certification from professional certification bodies (for transfers that meet specified thresholds).

Special regulatory rules

Beyond the foundational laws, specific regulations target AI applications:

• Three core AI-related regulatory instruments: the Internet Information Service Algorithm Recommendation Management Regulations (Algorithm Recommendation Regulations), the Internet Information Service Deep Synthesis Management Regulations (Deep Synthesis Regulations), and the Interim Measures for the Management of Generative Artificial Intelligence Services (Generative AI Interim Measures), form a complementary, multi-tier governance framework covering the full lifecycle of AI application and service provision. The Algorithm Recommendation Regulations lay the foundational rules for algorithm transparency, user rights protection and content moderation, mandating filing requirements for algorithms with public opinion attributes and prohibiting algorithm misuse. The Deep Synthesis Regulations govern AI-generated content, enforce user real-name authentication and mandatory labelling of synthetic text, images and audio to prevent disinformation. The Generative AI Interim Measures further refine standards for advanced generative AI models, emphasising the legality of training data, intellectual property protection, and content control to block illegal or harmful output.

• Specialised implementation rules: Under the comprehensive regulatory framework established by the three core AI-related regulatory instruments, there are supplementary specialised implementation rules. Currently, the key effective instrument is the Measures for Labelling AI-generated and Synthesised Content. It mandates that all AI-generated content, including text, images, audio and video, must be clearly labelled as synthetic to ensure that users can distinguish it from real content.

• National standards: China also implements AI-related national standards, which serve as a crucial bridge between legal provisions and practical implementation. These standards provide general technical guidelines and operational benchmarks for data and AI compliance.

Main scenarios of AI usage and development in China

Foreign companies in China typically engage with AI in two main ways, each with distinct compliance considerations.

• AI as a tool – internal office use for foreign companies as AI users

In this scenario, AI is employed to streamline internal operations, such as smart attendance systems, document processing and customer data analysis.

• AI in a product – external service provision by foreign companies as AI developers/service providers

Here, AI is embedded in products or services offered to external customers, like smart customer service systems, e-commerce recommendation engines or industrial inspection equipment.

Compliance considerations for foreign companies as AI users

When using AI internally, foreign companies should focus on several key compliance areas:

Data collection and processing

The internal use of AI tools inevitably involves the collection of personal information from employees and business partners. Foreign companies must ensure that all such data collection activities are fully compliant with Chinese law. Companies are required to clearly inform the relevant data subjects, namely their own employees and the employees of their business partners, about the purpose, scope, and intended use of the personal information being collected. Obtaining their explicit consent is a fundamental legal requirement under Chinese data law. Additionally, foreign companies should adopt the principle of collecting and processing only the minimum necessary personal information required to achieve their business functions, and they must ensure their employees and business partners’ employees are notified regarding any information collected and processed. For any sensitive data, such as employees’ biometric information like facial recognition data, or confidential commercial information from business partners, companies should, where possible, avoid collecting and processing it through AI tools. When handling such data is unavoidable, companies must employ anonymisation techniques to prevent unauthorised access and potential privacy breaches.

Data storage and export

Foreign companies must implement robust security measures for data storage, including encryption and strict access controls. These measures are vital to protect data from unauthorised access and potential cyber threats. Additionally, when using AI tools involves sharing data across borders, such as storing data directly on overseas servers or allowing access by a foreign headquarters, it is crucial for companies to comply with cross-border data transfer regulations. Depending on the type and volume of data, this includes conducting mandatory security assessments, obtaining certification for cross-border transfers, or having both domestic and foreign entities sign and submit standard contracts to the relevant regulatory authorities.

AI-generated content labelling

Chinese law requires that AI-generated content must be marked either explicitly or implicitly. Explicit marking involves adding indicators in the interface of synthetic content or interactive scenarios, presented as text, sound, graphics, etc., and clearly perceivable by users. Implicit marking involves embedding indicators in the data of synthetic content files through technical measures, which are not easily perceptible by users. Implicit marking is a legal obligation requiring AI tool providers to automatically embed indicators during the content generation process and to write them into the file metadata. For foreign companies using AI tools, if the generated content is for internal use only, they are generally exempt from explicit marking. In such cases, foreign companies should verify the implicit marking by the AI tool providers. If no implicit marking is found, they should request the provider to rectify it or complete it themselves. However, if foreign companies publish any AI-generated or synthesised content through online information and content dissemination services, or deliver it to external third parties, they are obligated to declare and label the content to ensure that it is recognisable as AI-generated. Additionally, they must update the implicit marking to maintain compliance.

Third-party AI tool procurement

When purchasing third-party AI tools, foreign companies should conduct thorough due diligence on suppliers’ data security qualifications. Service agreements should include clear data protection clauses that outline the responsibilities and obligations of both parties. A common scenario for foreign companies is that, based on the unified procurement management of multinationals’ headquarters, their Chinese affiliates may also deploy AI tools from overseas. Any AI tool launched in China must undergo the country’s algorithm filing and large model filing/registration processes. AI tools from overseas usually cannot meet such requirements. However, as an exception, current practices may not require overseas AI tools solely used internally to undergo the filing procedure, but companies will need to stay vigilant about regulatory changes.

Compliance considerations for foreign companies as AI developers / service providers

Developing AI for external use introduces additional compliance complexities.

Data training

During the data training phase, developers should use legal, diverse and objective data sources, filter out invalid or harmful data, and ensure that the training data is free from false, biased, or infringing content. For data involving personal information, developers should implement de-identification and other desensitisation processes to enhance privacy protection. They should also strengthen data security in key industry sectors, such as government affairs and finance, to guard against the leakage of critical data. In the development process, developers should standardise the data labelling process to improve accuracy and reliability.

Filing requirements

In China, all AI tools officially launched for operation, specifically AI services with public opinion or social mobilisation capabilities, as well as any AI tools made available to the public, must undergo the algorithm filing process and large model filing/registration. Additionally, filing with the cybersecurity multi-level protection scheme (MLPS filing) is also required. This constitutes the most critical compliance requirement for developers.

First, in accordance with the Algorithm Recommendation Regulations and the Deep Synthesis Regulations, the five categories of algorithms subject to mandatory filing—namely algorithms used for recommendations, generative and synthetic content, ranking and selection, retrieval and filtering, or scheduling and decision-making—essentially cover all core algorithm applications in internet information service scenarios. This means that virtually all enterprises engaged in AI development must first complete the algorithm filing process.

Second, under the Generative AI Interim Measures, entities that independently develop large models or provide generative AI services to the public after fine-tuning models developed by others must complete the large model filing process before their products are officially launched. In contrast, entities that directly use others’ large models without any secondary development only need to conduct large model registration. The latter process is relatively streamlined because the original developer of the AI foundation model has already completed the model’s filing, so the invoking company merely needs to register the fact that it is using the model for its own specific business scenarios.

A common scenario for foreign companies is that AI development conducted in China typically involves fine-tuning the open-source large models of other developers. For such companies, the required compliance steps include filing the algorithms corresponding to their products and completing the large model filing process. A key question arises here: Can the large models used for such development be sourced from overseas? As the underlying open-source large models, they must themselves have completed China’s algorithm and large model filing. In other words, if a foreign company in China uses an overseas open-source large model as the foundation, that model must have completed the relevant filing process in China. However, to date, no mainstream overseas AI models have passed China’s algorithm or large model filing processes, leaving foreign companies that rely on overseas models in a challenging position. Therefore, prioritising the adoption of domestic AI models is advisable.

Moreover, under the three foundational data laws, all network operators are required to undertake the MLPS filing. This obligation requires network operators to classify and grade their systems based on factors like service scope and data sensitivity. They must then complete compliance processes, including filing and level evaluation. During the filing process, a comprehensive review and assessment are conducted on various aspects of the company, including the establishment of data security management systems, technical security measures such as data classification and grading, the formulation of cybersecurity incident emergency plans, and the compliance of the entire process of personal information processing. The process concludes with the approval from the relevant public security department. For companies operating AI tools, this obligation is particularly important and is also explicitly reflected in some local administrative notices. Given the necessity of fulfilling the filing requirements, it is virtually impossible for foreign companies to consider deploying their systems overseas.

AI-generated content labelling

Foreign companies operating their own AI tools as service providers should add explicit or implicit markings to AI-generated content and deploy deepfake-detection tools in certain scenarios to verify the sources of information.

Data export

Data export is undoubtedly a key compliance concern for developers and operators of AI tools. It often involves large volumes of personal information or data from sensitive industries, which can trigger numerous additional compliance requirements. Therefore, without further elaboration, deploying AI-related systems overseas also presents practical difficulties that call for careful evaluation.

Outlook and suggestions

As China’s AI regulatory framework continues to evolve, data compliance has become a critical cornerstone and strategic necessity for foreign companies pursuing sustainable growth in the Chinese market. Looking ahead, amid advancing technology and increasingly refined regulations, foreign companies must remain vigilant and integrate data compliance into all business facets to tap into AI’s potential while ensuring legal and secure application. To proactively navigate this complex landscape, targeted measures are recommended as follows:

• Establish a dedicated compliance team to develop and implement tailored data compliance policies;
• Enhance technical safeguards such as data anonymisation and encryption to strengthen data security;
• Conduct regular employee training to improve compliance awareness and reduce inadvertent violations;
• Stay updated on I and data regulatory changes to adjust compliance strategies in a timely manner; and
• Proactively engage with regulatory authorities to gain insights and guidance, aligning business practices with regulatory expectations.

By implementing these measures, foreign companies should be well positioned to address and offset potential compliance risks and lay a solid foundation for stable development in China’s AI market.

---

Panpan Tang is a senior associate at CMS China.

As a top global law firm, CMS provides a full range of legal and tax services in over 50 countries, with more than 90 offices and 7,200 CMS lawyers worldwide. CMS China (Shanghai, Beijing, and Hong Kong) offers business-focussed advice tailored to companies’ needs.