Direct Answer / Overview

Foreign AI patent holders in China face a grant rate of approximately 40–45%, compared to roughly 55–60% for domestic applicants, a gap that underscores the importance of navigating China’s patent system with precision. China’s patent regime, anchored by the amended PRC Patent Law (专利法, zhuānlì fǎ) which took effect on June 1, 2021, offers substantive protection for foreign AI inventions — provided applicants understand the unique eligibility criteria, filing obligations, and enforcement mechanisms that distinguish the Chinese system from other major jurisdictions. This article explains how foreign AI innovators can protect their intellectual property under China’s current legal framework.

Direct Answer / Overview

Yes, China’s patent laws protect foreign AI inventions — but the level of protection depends heavily on how the invention is claimed and prosecuted. The PRC Patent Law, as amended in 2020 and effective June 1, 2021, recognizes three patent types: invention patents (发明专利, fāmíng zhuānlì), utility model patents (实用新型, shíyòng xīnxíng), and design patents (外观设计, wàiguān shèjì). For AI inventions, the most relevant category is the invention patent, which undergoes substantive examination and can receive up to 20 years of protection from the filing date.

The governing body is the China National Intellectual Property Administration (国家知识产权局, Guójiā Zhīshì Chǎnquán Jú, or CNIPA). CNIPA’s revised Examination Guidelines for Patent Applications, published in December 2019 and effective February 2020, explicitly address the patentability of artificial intelligence. Under these guidelines, AI algorithms are considered patent-eligible when they are integrated with a “technical solution” — meaning the algorithm must be applied to a specific technical problem and produce a concrete technical effect. This represents a significant shift from earlier practice, where pure algorithms were routinely rejected as abstract methods.

Foreign AI companies should also be aware that China’s patent system operates on a first-to-file basis. There is no grace period for public disclosures made before filing (unlike in the United States), meaning any prior publication or public use of the AI invention before filing can destroy novelty. Foreign applicants must also comply with Article 18 of the PRC Patent Law, which requires any applicant without a business address in China to use a CNIPA-registered patent agency. As of 2025, approximately 1,200 such agencies are registered with CNIPA.

Patent Protection Framework for AI Inventions

China’s patent protection for foreign AI inventions operates across three core legal instruments: the PRC Patent Law itself (as amended in 2020), the Implementing Regulations of the PRC Patent Law, and the CNIPA Patent Examination Guidelines. Together, these documents define what constitutes a patentable invention, the scope of protection, and the remedies available for infringement.

Under Article 42, paragraph 2 of the amended Patent Law, patent term compensation is available to remedy unreasonable examination delays. Specifically, if a patent is granted more than four years from the filing date and more than three years from the date of examination request, the patent term may be extended. This provision is especially relevant for foreign AI applicants, whose applications tend to undergo longer examination periods due to the complexity of AI-related claims and the need for technical expert review at CNIPA.

Additionally, Article 48 of the Patent Law permits patent holders to contribute their intellectual property as capital investment (知识产权出资, zhīshì chǎnquán chūzī) in Chinese companies. This requires an independent valuation of the patent and can be a strategic tool for foreign AI companies establishing joint ventures or subsidiaries in China.

Patent Eligibility of AI Inventions Under CNIPA Guidelines

The patent eligibility of AI inventions under Chinese law hinges on whether the invention qualifies as a “technical solution” (技术方案, jìshù fāng’àn). Under the CNIPA Examination Guidelines revised in December 2019, inventions that involve artificial intelligence algorithms are patentable subject matter if they meet two core criteria:

1. Integration with a technical field: The AI algorithm must be applied to a specific technical problem — for example, image recognition in medical imaging, natural language processing for machine translation, or predictive modeling for industrial process control. Pure abstract algorithms or mathematical methods that lack a technical application are excluded under Article 25 of the Patent Law.
2. Production of a technical effect: The invention must produce a concrete, technical result that goes beyond mere data processing. For example, an AI model that improves the accuracy of a diagnostic system, reduces latency in an autonomous vehicle’s decision-making, or optimizes energy consumption in a data center would qualify. The effect must be demonstrable and measurable within the relevant technical domain.

The 2020 guidelines specifically added a new section titled “Inventions Related to Artificial Intelligence” to Chapter 9 of Part II. This section clarifies that AI-based inventions are not per se excluded from patentability, even though they may involve algorithms or mathematical methods. The key legal basis is found in the guide’s interpretation of Article 2.2 of the Patent Law, which defines an “invention” as “a new technical solution proposed for a product, process, or improvement thereof.”

It is important to distinguish between patent protection and other forms of IP protection for AI inventions. While algorithms integrated with technical solutions can be patented, pure AI models — including trained neural network weights, training datasets, and model architectures — may also be protected as trade secrets under Article 9 of the Anti-Unfair Competition Law (反不正当竞争法, fǎn bù zhèngdàng jìngzhēng fǎ). Chinese courts have recognized that AI models meeting the criteria of “commercially valuable and subject to reasonable confidentiality measures” qualify as trade secrets. For many foreign AI companies, a dual strategy of patenting the technical application while keeping the core model architecture and training data as trade secrets is the most effective approach.

One notable challenge in AI patent eligibility is the treatment of inventions that rely on deep learning and large language models. CNIPA examiners often scrutinize whether the contribution lies in the technical application or merely in the data processing pipeline. Foreign applicants should ensure the patent claims explicitly describe how the AI algorithm solves a technical problem and include details of the technical effect, such as improved processing speed, reduced computational resource usage, or enhanced accuracy in a specific industrial application.

Filing Requirements for Foreign Applicants

Foreign AI companies seeking patent protection in China must navigate a set of procedural requirements that differ significantly from other major jurisdictions. The most important of these is set out in Article 18 of the PRC Patent Law, which states: “Where a foreign applicant who does not have a regular residence or place of business in China files a patent application in China, the application shall be handled by a patent agency that is registered with CNIPA.”

Key filing requirements for foreign AI applicants include:

• Mandatory patent agency representation: Foreign applicants without a Chinese business address must appoint a CNIPA-registered patent agency. There are roughly 1,200 such agencies as of 2025, of which only a subset specialize in AI and fintech-related patents. Choosing an agency with technical expertise in AI — and the ability to communicate effectively in English and Chinese — is critical to application quality.
• Certified priority documents: If claiming priority from a foreign filing (e.g., a provisional or non-provisional application in the United States, Europe, or Japan under the Paris Convention or PCT), the foreign applicant must submit a certified copy of the priority application within 16 months of the priority date.
• Chinese-language specification: All patent applications must be filed in Chinese. If the original specification was prepared in English, a certified Chinese translation must be submitted. Mistranslation of technical terms — especially AI-specific concepts like “neural network” (神经网络, shénjīng wǎngluò), “machine learning” (机器学习, jīqì xuéxí), or “deep learning” (深度学习, shēndù xuéxí) — can lead to claim scope narrowing or rejection.
• Substantive examination request: Invention patents require a substantive examination request within three years of the filing date. Foreign applicants should file this request promptly to avoid delaying the examination process unnecessarily. Combined with the patent term compensation provisions of Article 42(2), strategic timing of the examination request can affect the overall patent term.
• Security review for sensitive technologies: Under the Patent Law and the Export Control Law, AI inventions in certain fields (e.g., cryptography, dual-use AI technologies, large-scale data analytics) may require a security review before a foreign filing can proceed. If the invention was made in China, a domestic filing (or security clearance) is required before any foreign filing can occur.

Foreign applicants should budget approximately RMB 5,000–15,000 for official filing fees, plus attorney and translation costs ranging from RMB 30,000–80,000 for a typical AI invention patent application, depending on complexity. The examination timeline for AI-related inventions at CNIPA averages 22–36 months from the filing date, though this can extend longer for applications involving complex algorithm claims.

Enforcement: Damages, Injunctions, and Trade Secrets

Enforcing AI patents in China has become significantly more viable following the 2020 amendments to the PRC Patent Law. The most impactful change is found in Article 71, which governs patent infringement damages.

Under Article 71, courts may award statutory damages of up to RMB 5 million (approximately USD 690,000) for patent infringement where actual losses, profits, or reasonable royalties cannot be adequately calculated. In cases of willful infringement, the court may award damages of up to five times the actual loss or infringer’s profit. This punitive multiple is a substantial deterrent and brings China’s enforcement regime closer to the standards found in the United States and Germany.

The 2020 amendment also introduced several pro-patentee provisions:

• Reversal of burden of proof: For process patents, the court may shift the burden of proof to the alleged infringer to demonstrate that their process is different — a significant advantage for foreign patent holders where process details are difficult to obtain.
• Preservation of evidence: Courts may order preservation of evidence, including access to financial records and manufacturing facilities, before a final ruling on the merits. This is particularly useful in AI patent disputes where the infringement may involve software code or model architectures that can be easily concealed.
• Preliminary injunctions: Courts may issue preliminary injunctions where there is a likelihood of irreparable harm and the patent is shown to be valid and infringed. While Chinese courts historically granted injunctions less frequently than their U.S. counterparts, recent practice — especially in specialized IP courts in Beijing, Shanghai, and Guangzhou — has shown an increasing willingness to grant injunctions in cases involving clear infringement of high-value patents.

Beyond patent enforcement, foreign AI companies can also rely on trade secret protection under Article 9 of the Anti-Unfair Competition Law. The law defines trade secrets as “technical information and business information that is unknown to the public, has commercial value, and for which the rights holder has taken reasonable confidentiality measures.” AI models, including trained model parameters, training datasets, and model architectures, can qualify as trade secrets if properly protected.

Chinese courts have handled several high-profile AI trade secret cases in recent years. In 2022, a Chinese court awarded damages of RMB 10 million (approximately USD 1.4 million) in a case involving the theft of a facial recognition AI algorithm, finding that the algorithm’s source code and trained model constituted trade secrets under Article 9. This demonstrates that trade secret enforcement can be an effective complement — or alternative — to patent enforcement for AI inventions, especially where the core value lies in the model itself rather than its technical application.

Practical Filing Strategy for Foreign AI Companies

Based on the legal framework and enforcement landscape described above, foreign AI companies should consider the following strategic approach to patent protection in China:

1. File early and file in Chinese. China is a first-to-file jurisdiction with no grace period. File a PCT or Paris Convention application well before any public disclosure, and invest in a high-quality Chinese translation prepared by a patent translator with AI/technical expertise. A single mistranslation of a key claim term can permanently narrow the scope of protection.
2. Select a specialist patent agency. Of the approximately 1,200 CNIPA-registered agencies, focus on those with demonstrated experience in AI and fintech patent prosecution. Ask for their grant rate for foreign AI patent applicants — this is a useful benchmark. Agencies based in Beijing (near CNIPA headquarters) tend to have the most interaction with examiners.
3. Structure claims around technical effect. Ensure each independent claim explicitly describes the technical problem solved and the technical effect produced. Avoid overly abstract or function-based claiming. Claims that recite specific improvements in processing speed, accuracy, or resource utilization are more likely to survive examination.
4. Leverage patent term compensation strategically. Article 42(2) can extend the patent term by the period of examination delay beyond four years from filing and three years from the examination request. If CNIPA examination takes longer than expected, track the dates carefully and file a term compensation request upon grant.
5. Maintain strict trade secret protection for core AI assets. Patent filings require public disclosure, which may not be desirable for core AI models and training datasets. Use patents to protect the technical application and system architecture, while keeping the actual trained model weights and proprietary training data as trade secrets protected by confidentiality agreements, access controls, and encryption.
6. Consider dual filing (invention + utility model). For AI-related hardware innovations (e.g., specialized AI chips, sensors, edge computing devices), consider filing both an invention patent and a utility model patent on the same date. Utility model patents are granted without substantive examination (typically within 6–12 months) and provide early provisional protection while the invention patent undergoes full examination. However, the same invention cannot be protected by both a utility model and an invention patent simultaneously — one must be surrendered if both are granted.

The grant rate gap between foreign and domestic AI applicants (40–45% vs. 55–60%) is narrowing as CNIPA examiners gain more experience with AI-related claims. Foreign applicants who invest in well-drafted specifications, clear technical-effect language, and experienced local representation can achieve grant rates at the higher end of that range. Moreover, China has become one of the most active patent filing jurisdictions for AI globally — in 2023, CNIPA received over 130,000 AI-related patent applications, accounting for roughly 40% of all AI patent filings worldwide. This volume creates both opportunity and competition. Foreign AI companies that enter the system with a well-prepared strategy are well-positioned to secure meaningful protection in the world’s second-largest AI market.

Where to Go From Here

Based on what you just read:

• Ready to act? Read [guide: SLUG-TO-BE-FILLED]
• Still comparing? See [comparison: SLUG-TO-BE-FILLED]
• Need numbers? Try [tool: SLUG-TO-BE-FILLED]

— China Gateway 360 —
Remote China market entry support, built around execution.