How Do China’s Patent Laws Protect Foreign AI Inventions in 2024?
China’s patent system provides a robust, yet highly specific, framework for protecting foreign artificial intelligence (AI) inventions. However, the path to securing a 发明专利 (Invention Patent, fāmíng zhuānlì) is not identical to that in the US or Europe. To qualify for protection under the 国家知识产权局 (China National Intellectual Property Administration, CNIPA, guójiā zhīshì chǎnquán jú), an AI invention must constitute a “technical solution.” Currently, foreign applicants hold roughly 15% of the nearly 180,000 active AI patents in China, but face a grant rate of only 65%—significantly lower than the 85% average for traditional mechanical inventions. This gap exists because many foreign applications fail to meet the strict “technical solution” criteria defined under China’s Patent Law. This FAQ explains exactly how foreign companies can navigate these rules to effectively protect their AI IP.
The Core Challenge: Defining a “Technical Solution” for AI
The most critical concept for foreign AI inventors to understand is Article 2.2 of China’s Patent Law, which defines a patentable invention as a “technical solution.” This is China’s version of subject-matter eligibility, but applied very strictly.
The CNIPA generally views pure algorithms, mathematical methods, and business rules as abstract mental activities, which are explicitly not patentable. To overcome this, your patent application must demonstrate a 技术方案 (technical solution, jìshù fāng’àn) that solves a specific technical problem using technical means and produces a tangible technical effect.
For example, if your AI invention is a new recommendation algorithm for an e-commerce platform, the CNIPA might reject it as a “business method.” However, if you frame the same invention as solving a specific *technical* problem—such as “reducing server load by optimizing data query pathways”—it suddenly becomes patentable. The key is in the framing and the technical context of the application.
Decision Framework: Patent vs. Trade Secret for AI
Foreign executives must decide early whether to pursue a patent or rely on 商业秘密 (Trade Secret, shāngyè mìmì) protection. This decision is not trivial. China has a “first-to-file” system, meaning if you delay filing, a competitor could patent your invention first.
If your AI innovation: Directly improves a physical device, manufacturing process, or specific technical field (e.g., autonomous driving obstacle detection, medical image diagnosis), choose a standard Invention Patent. The technical context is easy to prove.
If your AI innovation: Is a unreleased, highly valuable core algorithm used internally (e.g., a unique trading algorithm or a proprietary data processing pipeline), choose a robust Trade Secret strategy. This avoids the public disclosure required by patents and the risk of an office action rejection.
Cost: Loss of trade secret protection globally if the patent is rejected or invalidated. Potential value: RMB 1,000,000+ (lost competitive advantage).
Fix: Work with a CNIPA-specialized attorney to write the specification using technical flowcharts and general implementations, keeping the core algorithm a trade secret.
Comparison Table: AI Patent Eligibility (CNIPA vs. USPTO)
| Criteria | China (CNIPA) | USA (USPTO) |
|---|---|---|
| Subject Matter Exclusion | Abstract ideas, mathematical methods, business methods are strictly excluded unless they form a “technical solution.” | Laws of nature, natural phenomena, and abstract ideas are excluded. “Alice” test is applied. |
| Technical Requirement | High. Must solve a specific technical problem using technical means. A computer merely implementing a rule is not enough. | Moderate. Must be “significantly more” than an abstract idea. Integration into a practical application is key. |
| Data as Invention | Purely data generation models (GANs) are harder to patent if not linked to a specific technical use. | Easier to patent data generation methods if they demonstrate technological improvement. |
| Grant Rate | ~65% for foreign AI applications (2023 estimate). | ~55-60% for AI software applications (USPTO data). |
Context: While the USPTO is slightly more lenient on software *per se*, the CNIPA is more forgiving on technical description detail. A well-drafted specification in China that clearly states the “industrial applicability” has a strong chance of grant.
Timeline and Cost of AI Patent Protection in China
Securing a patent in China is not an overnight process. Understanding the typical timeline helps in strategic planning.
- Average Pendency: 22 to 36 months (about 2-3 years) for a standard AI invention patent application. Using the Patent Prosecution Highway (PPH) can reduce this to 12-18 months if a corresponding US or Japanese patent has already been allowed.
- Filing Costs: Typical attorney fees for drafting, translating, and filing an AI patent application in China range from RMB 20,000 to RMB 40,000 (approx. $2,800 – $5,600 USD). This is lower than US filing costs.
- Accelerated Examination: If your AI invention relates to a “National Key Technology” (e.g., AI for public health, green tech), you may qualify for an accelerated examination, reducing wait time to 6-12 months.
- Litigation Costs: Enforcing a patent in China is cheaper than in the US. Standard litigation costs range from RMB 200,000 to RMB 500,000. High-stakes cases involving major AI algorithms can exceed RMB 1,000,000.
Cost: A rejection or office action requiring narrow re-drafting. Cost in legal fees and delays: RMB 15,000 – RMB 50,000.
Fix: Always conduct a CNIPA-specific prior art search with a local Chinese patent agent who specializes in AI before filing.
AI Inventorship and Data Ownership
A frequently asked question is whether an AI can be listed as an inventor. In China, as in most jurisdictions, the answer is no. The concept of “inventor” under the Patent Law is explicitly limited to a natural person. While DABUS (the AI system) has been the subject of litigation globally, CNIPA has firmly rejected its designation as an inventor.
More critically for foreign firms is the issue of data ownership. Most AI patents rely on training data. If your AI invention is developed using data generated in China (e.g., from Chinese users or operations), you must ensure you have the proper data rights and privacy compliance (PIPL and CSL). If the data is improperly sourced, the patent itself could be challenged on the basis that the “technical contribution” was obtained through illegal means.
Cost: Abandonment of the patent application. Loss of all sunk costs (RMB 20,000+ filing fees) and critical time-to-market protection.
Fix: Ensure your local counsel drafts responses strictly proving the “technical problem-solution” framework. Never argue commercial success; argue technical necessity.
Next Steps for Protecting Your AI in China
Navigating CNIPA’s strict technical requirements demands a tailored strategy. Based on the risks and decision framework above, here are your actionable next steps:
- Audit Your AI Portfolio: Identify which of your AI inventions have strong “technical context” (e.g., hardware integration) vs. pure algorithm. Read our guide: AI Patentability Audit for Foreign Firms.
- Draft with Chinese Rules in Mind: Do not simply translate a US application. It must be re-framed to meet the technical solution standard. Review our CNIPA AI Drafting Checklist.
- Evaluate Trade Secrets: For core algorithms that cannot be easily reverse-engineered, a trade secret strategy might offer indefinite protection without disclosure. Compare Trade Secret vs. Patent for China Tech.
— China Gateway 360 —
Remote China market entry support, built around execution.
