A Guide to Patent Eligibility of AI-Generated Inventions in China for Foreign Companies

With the rapid development of artificial intelligence technology, an increasing number of technical solutions generated autonomously or with assistance by AI systems are appearing in patent applications. For foreign companies seeking patent protection in the Chinese market, understanding the examination standards of the China National Intellectual Property Administration (CNIPA) for AI-generated inventions is crucial. This article provides a comprehensive operational guide for foreign applicants from three perspectives: legal framework, examination practice, and practical strategy.

I. Legal Framework for AI-Generated Inventions Under Chinese Patent Law

China's current Patent Law does not contain specific provisions targeting AI-generated inventions, but it establishes the fundamental principle that "inventions" must be completed by "natural persons." The fourth amendment to the Patent Law in 2020 elevated "intellectual property protection" to a new height, while also adding specialized provisions in the Examination Guidelines for emerging technology fields such as artificial intelligence and big data.

According to the Implementing Regulations of the Patent Law and the Patent Examination Guidelines (2023 Revision) issued by CNIPA, the core criteria for determining the patent eligibility of AI-generated inventions do not fundamentally differ from those for traditional inventions. The following three conditions must still be satisfied:

1. Subject matter eligible for patent protection — it must constitute a technical solution, not a pure intellectual activity rule or mathematical method
2. Novelty — it must not form part of the prior art
3. Inventive step — compared with the prior art, it must possess prominent substantive特点和显著的进步

However, AI-generated inventions face a fundamental challenge in practice: the determination of inventor identity. Chinese patent law requires that an inventor must be "a natural person who has made creative contributions to the substantive features of the invention." When an invention is generated by an AI system, the AI itself cannot be listed as an inventor, and the applicant must identify a natural person who can be legally recognized as the "inventor."

II. Boundaries of Patentable Subject Matter for AI-Generated Inventions

When examining AI-related patent applications, CNIPA first determines whether the claims fall within the definition of a "technical solution" under patent law. This is a critical threshold that foreign applicants most easily overlook.

According to the Patent Examination Guidelines, pure algorithms, mathematical models, business rules, and methods do not constitute technical solutions. The specific criterion is: whether the claim employs technical means, solves a technical problem, and produces a technical effect.

For AI-generated inventions, the following scenarios are generally considered patentable subject matter:

• AI-assisted generation of specific product designs: such as mechanical structures, circuit layouts, or architectural designs automatically generated by AI, provided the final product is practical and manufacturable. For example, a heat sink fin structure generated by an AI system through reinforcement learning, though designed by AI, ultimately resides in a specific physical product.
• AI-optimized process flows: such as chemical reaction conditions, material composition ratios, or manufacturing parameter combinations generated by AI, provided these processes have a clear technical effect and can be reproducibly implemented.
• The AI training method itself: if the claim protects the training method of an AI model and that method employs improved technical means (such as a novel loss function design, optimization of attention mechanisms, etc.), it is generally considered a technical solution.

The following scenarios are more likely to be deemed unpatentable subject matter:

• Purely data classification or prediction methods: those relying solely on mathematical operations to classify or predict data, without being tied to a specific technical field.
• AI-generated business strategies: such as marketing plans or pricing strategies generated by AI, which fall under the rules of mental activity.
• AI outputs lacking a description of practicability: claims that merely describe the input and output of AI without disclosing the technical path from input to output.

III. Sufficiency of Disclosure and Reproducibility Requirements

When filing applications for AI-generated inventions, foreign companies must pay particular attention to the requirement of "sufficiency of disclosure." Article 26, Paragraph 3 of China's Patent Law provides that the description shall set forth the invention in a manner sufficiently clear and complete so that a person skilled in the relevant technical field can carry it out.

For AI-generated inventions, "sufficiency of disclosure" faces unique challenges:

• The basic architecture of the AI system and the source of training data
• The specific input conditions or prompts used to generate the invention
• The "human intervention" steps taken to filter and confirm the AI output
• At least one specific embodiment that can be reproducibly implemented

IV. Distinguishing AI Contribution from Human Contribution in Inventive Step Examination

Inventive step examination is the area of greatest uncertainty for AI-generated inventions. When examining, CNIPA focuses on whether the invention possesses "prominent substantive features" compared with the prior art.

When AI serves merely as a tool assisting humans in completing an invention, the standard for determining inventive step is generally consistent with that for traditional inventions. The examiner uses the "person skilled in the art" as the benchmark, but here the "person skilled in the art" is presumed to possess the general ability to use AI tools. This means:

• If an invention is merely the result of running an existing AI model in a routine manner, the examiner may deem it as something a person skilled in the art could obtain through "routine experimentation," thereby denying inventive step.
• If the invention involves an innovative use of the AI model, a special training strategy, or the creative integration of domain knowledge, it is more likely to be found to possess inventive step.

For inventions with a high degree of AI autonomy and relatively little human contribution, foreign applicants should proactively elaborate in the application documents on the human creative contributions in the following areas:

1. Creativity in the problem-definition phase: how a human identified and defined a technical problem, and whether the problem itself is innovative
2. Creative construction of training data: whether the selection, cleaning, and annotation of training data involved human creative labor
3. Creative design of AI architecture: whether the AI model was customized or adapted for a specific technical field
4. Creative judgment in output selection: how a human creatively selected and combined from multiple candidate solutions generated by AI
5. Creative work in verification and improvement: how a human tested the AI-generated solution and made improvements and optimizations

V. Practical Operational Recommendations for Foreign Companies

Based on the above analysis, when filing patent applications for AI-generated inventions in China, foreign companies are advised to adopt the following practical strategies:

Ensure that the inventors include natural persons who made substantive manual modifications or creative selection and confirmation of the AI output. Merely providing "instructions to use AI" does not constitute inventor qualification.

• In the background section of the specification, describe the technical bottlenecks of traditional methods
• In the summary of the invention section, emphasize how humans designed the AI workflow and defined the technical boundaries of the problem
• In the detailed description section, explain in detail how humans selected, verified, and modified the final technical solution from the AI output

• Provide data comparing the effects of the AI-generated solution with traditional solutions
• Demonstrate that the technical solution generated by AI produced unexpected technical effects
• For structural innovations in the AI model itself, provide experimental evidence for parameter selection

• The first application can focus on the parts with stronger human creative contribution
• Gradually expand, through divisional applications, to AI-generated but human-verified technical solutions
• Use the priority system to secure more preparation time

When CNIPA issues an Office Action involving issues of insufficient disclosure or lack of inventive step related to AI, the following response approaches are recommended:

• Submit supplementary explanations of the AI workflow
• Argue that the AI's role in the invention is as a "tool" rather than an "inventor"
• Cite relevant case law from Chinese courts or the Patent Reexamination Board concerning AI inventions
• Narrow the scope of claims when necessary, excluding the more controversial, fully AI-generated parts

VI. Risk Warnings and Future Outlook

Currently, China's legal framework regarding the patent eligibility of AI-generated inventions remains in an exploratory phase. Although CNIPA has added AI-related provisions to the Examination Guidelines, its attitude toward fully AI-generated inventions is trending cautious.

Foreign companies should be aware of the following risks:

• Post-grant invalidation risk: Patents that overly rely on AI-generated content without substantial human contribution may be invalidated in subsequent invalidation proceedings. It is recommended to maintain clear records of human creative contributions in the patent portfolio.
• Consistency across cross-border applications: China's examination standards for AI inventions differ from those in the United States and Europe. The same invention may face different grant outcomes in different jurisdictions. It is recommended to adjust the application strategy specifically for the Chinese market.
• Extended examination periods: Patent applications involving AI-generated inventions may require longer examination cycles at CNIPA. Proper timeline expectations should be managed.

In the long term, China is actively researching a legal protection framework for AI-generated content. In 2023, the Beijing Internet Court issued a landmark decision in an AI-generated image copyright case, recognizing that works created by humans using AI tools can, under certain conditions, receive copyright protection. This trend may gradually extend to the patent field.

Foreign companies should continuously monitor CNIPA's examination developments and judicial practice, while maintaining communication with professional Chinese intellectual property agencies (such as WeRights) to ensure that their filing strategies stay aligned with the latest examination standards. Should you have any questions, feel free to contact us via Telegram @token_1_com.