Securing AI Patents in China: The Regulatory Framework

China has been a powerhouse in AI development and deployment - since 2013, China has overtaken the US as the country with the most number of AI-related patents granted (within all jurisdictions).[1] To continue fostering this surge of innovation, China has already taken significant steps in this direction with the release of its New Generation Artificial Intelligence Development Plan (2017), a comprehensive plan outlining goals for the development and implementation of AI by 2030.  In addition to this strategic plan, China has also made crucial adjustments to its regulatory framework. The 2019 revision of the China National IP Administration (CNIPA) Guidelines for Examination included notes about the patentability of AI inventions, marking a significant step towards clarifying the legal landscape for AI-related patents.

The lawmakers in the People’s Congress are discussing a national AI Law.[2] China is continuing to refine its regulatory landscape to balance innovation with ethical and societal considerations, and China’s increasing focus on intellectual property rights and innovation especially for AI inventions makes China an attractive hub for companies to protect their inventions.

I. Chinese patentability criteria

Chinese patent law establishes that inventions must involve technical means, solve technical problems, and achieve technical effects to be patentable. Article 25(2) of the Patent Law excludes the patentability of rules and methods of mental activities,  whereas Section 2, Chapter 9, Article 6.1.2 of the Guidelines for Examination (2023 revision) delineates that while abstract ideas and mental activities are excluded from patentability, claims that incorporate technical features—such as those related to computer programs, algorithms, and artificial intelligence (AI)—may be patentable if they demonstrate a tangible technical contribution and effect.

Patent Law Article 25

“No patent right shall be granted for any of the following:

…

(2) rules and methods for intellectual activities

…”

Guidelines for Examination Section 2, Chapter 9, Article 6.1.2

“When examining whether a claim that includes algorithmic features or features of business rules and methods belongs to a technical solution, it is necessary to consider all the features described in the claim as a whole. If the claim describes the use of technical means that utilize natural laws to solve a technical problem, and as a result, achieves a technical effect that conforms to natural laws, then the solution defined by the claim belongs to the technical solution as described in the second paragraph of Article of the Patent Law.”

The key aspects that define a patentable technical solution under Chinese law are the use of technical means that utilize natural laws to solve a technical problem, resulting in a technical effect that conforms to natural laws. For algorithmic claims specifically, the guidelines require that if the steps involving algorithms are closely related to the technical problem and process data with technical significance, and if the execution of the algorithm directly reflects the process of solving a technical problem, resulting in a technical effect, the solution could be considered a technical solution.

Zooming in on AI and big data algorithms in particular, inventions involving improvements in AI algorithms (deep learning, classification, clustering, etc.) may qualify as technical solutions if they solve technical problems such as enhancing hardware computation efficiency or execution effectiveness. 

Novelty and inventiveness are evaluated on all features described in the claim, including both the technical and non-technical features (algorithms). Evaluation of novelty and inventiveness is outlined in Section 2, Chapter 9, Article 6.1.3 of the Guidelines for Examination.

Guidelines for Examination Section 2, Chapter 9, Article 6.1.3

“When conducting a novelty examination of an invention patent application that involves algorithmic features of business rules and methods, all the features described in the claims should be considered. These features include both the technical features and algorithmic features or business rules and methods features.”

The practical application of these principles can be observed in various sorts of patent applications and rejections. For example, example 11 of Guidelines for Examination Section 2, Chapter 9, Article 6.2 discusses a patent application that proposes a method for detecting the fall state of humanoid robots. The technology is based on multi-sensor information that includes real-time fusion of robot gait phase information, posture information, zero-moment-point information, and utilizes a decision system to determine the current stability and controllability of the robot which thus provides a reference for the robot’s next action.

According to the claims document, the solution effectively improves the stability state of the robot and the reliability and accuracy of predicting its potential fall direction—a demonstration of functional interactions between algorithm features and technical features. The algorithm of the decision-making and its application to judging the robot’s stability state (its technical character) do not belong to general knowledge in the field, so it possesses novelty. This application would therefore be granted.

Applications involving novel AI techniques or big data processing methods are rigorously examined to ensure they demonstrate technical contributions and are not merely abstract ideas. The ability to link the technical implementation of an algorithm to a tangible improvement or application-specific outcome is crucial for patent eligibility.

The guideline is relatively new, and there is limited case law to illustrate how CNIPA enforces these laws in practice. Examining current USPTO and EPO practices can provide insights into the future direction of Chinese examination procedures.

II. EPO Patentability Criteria

Patentability through the European Patent Office (EPO) is generally given by the European Patent Convention (EPC) Article 52. Computer programs, AI, and algorithms hinge on Articles 52(2) and 52(3), which exclude patenting of mathematical models, algorithms, and computer programs as such. However, AI is generally patentable as a subgroup of Computer Implemented Inventions (CIIs)[3].

Like the US, due to the nature of computer programs, models, and/or algorithms, they are seen as mathematical procedures, or abstract activities that do not involve a technical character. A 2019 revision of the EPO Guidelines for Examination includes a specific section on computer programs, Artificial Intelligence, and Machine Learning. These guidelines explain additional technical characteristics that could make a computer program patentable.

Per the EPO Guidelines for Examination, claims involving technical means (i.e.: hardware components, technical processes) or directed to a device are considered to have technical character. “Technical character as a whole” implies that even if parts of the claim involve non-technical aspects, the presence of technical means can render the claim patentable. For AI/ML specifically, merely training an algorithm with data is insufficient to confer technical character; the technical contribution must come from how the algorithm is applied or technically implemented.

Case Law states that novelty and inventiveness are evaluated on only technical characters, which further underscores the need to clearly define the technical character in the claims. Non-technical features are examined for their contribution to the overall technical character.

T0697/17 regarding Microsoft’s patenting of SQL Extensions exemplifies EPO’s evaluation of computer-implemented inventions. The application was initially rejected by the EPO examining division for a lack of technical character and inventive step, but since the application was found to use non-technical features to solve a technical problem, it is legitimate to establish a technical contribution of the application overall. This case clarifies and reinforces the standards and serves as a reference point for applicants and examiners in understanding the nuanced requirements for patent eligibility of CIIs.

While the standards have remained relatively clear and constant through the decades it has been implemented, the requirements for AI technologies remain stringent and challenging to meet. The strict yet consistent framework demands that applicants provide substantial acceptable technical details to navigate the nuance requirements for patent eligibility in AI innovations.

III. USPTO Patentability Criteria

General patentability criteria in the United States stem from Title 35 of the US Code Section 101 defining the subject matter for which patents may be obtained, and Section 102, which defines statutory novelty and other conditions for patentability.

While eligibility of the subject matter is defined by the four broad statutory categories of invention (process, machine, manufacture, composition of matter), patentability is defined by Section 102 of whether the invention is new, useful, and non-obvious. Generally, laws of nature, natural phenomena, and abstract ideas are considered to be exceptions of § 101 and deemed unpatentable.

To compensate for claims that do have a factor of the judicial exceptions to the § 101 as mentioned before, the Mayo/Alice eligibility test was established in 2014. This test is established by two major Supreme Court decisions—Mayo v. Prometheus Laboratories established step 2A, determining whether a claim is directed towards a judicial exception, and if so, whether it contains sufficient inventive concept to overcome judicial exception (Alice v. CLS Bank International, step 2B)

However, this test that seems straightforward has proven difficult to implement consistently, especially for software and algorithm patents[4]. Algorithms, by their nature as a series of mathematical steps and procedures, fall under the abstract ideas category. The contours of this eligibility test remain unclear even after several years of successive decisions that have progressively driven the test to be more stringent, with previous decisions having enlarged the breadth of ineligible subject matter.

For example, in 2014, I/P Engine Inc. v. AOL Inc., a patented search advertising algorithm was invalidated.  Judge Mayer explained the verdict by issuing a concurrence that interprets the test as a “technological arts” test that mandates advances to be significant and well-defined: claims cannot be “overly broad” in proportion to the technological dividends they yield. This case demonstrates that there is an increasingly limiting patent eligibility space despite minimal changes to the actual wording. Interpretation of the test differs with different judges or different courts. If any sort of view regarding algorithms and LLMs arises from the courts it may be less due to changes from Alice/Mayo and more a result of the changing views of the judges.[5]

The subjective nature of the Alice/Mayo test and its significant uncertainty and variability outcomes have made it difficult for innovators to secure patents for AI algorithms and other software innovations. The evolving legal landscape continues to influence the patent eligibility space, with judicial interpretations playing a critical role in shaping the boundaries of what is considered patentable. As such, the future of AI patenting in the US will likely depend on further clarifications from the courts and potential legislative adjustments to better accommodate the unique challenges posed by rapidly advancing technologies.

IV. Comparison

The patentability for AI and algorithms in the US, Europe, and China share some similarities but also exhibit notable differences.

Table I: Notable Differences

A sample patent of a software robot for programmatically controlling computer programs to perform tasks filed in both USPTO and EPO demonstrates how differences (highlighted or underlined) in claim wording can appeal to each system.

In the very first line of the US patent claim 1, additions like “controlling multiple application programs” emphasize the flexibility of the patent, allowing it to broaden the scope and cover various system configurations. In the EP application, phrases like “a software robot computer program” and “cause at least one hardware processor to perform: accessing the software robot computer program;”  indicate the use of a specialized software component (technical specificity) to update an object state (highlighting a technical process) and explicitly describes an interaction between the hardware processor and the software robot computer program (what is defined as “technical character”).

The US claim also adds detailed procedural steps of the program itself, ensuring the invention is viewed as a specific, technical application rather than a general abstract idea. The detailed procedural steps can also demonstrate non-obviousness with the USPTO—the EP claim does not include this amount of procedural detail because its evaluation of inventiveness is based on contribution to technical character rather than detailed procedural steps.

The US patent claim saves on details of the technical effect, which the EP patent claim includes because the EPO could use this detail to evaluate (1) the technical character, (2) the details of this technical problem-solving, and (3) the inventiveness of this technical process. In place of this, the US patent claim chooses to include a blurb about control over a second application system because it could show that the system is versatile and capable of handling complex tasks involving multiple applications. The EP claim did not describe the “second application program” because the technical performance of the first application program is its core inventive aspect: including the second application program in the same amount of detail may make it hard to defend as a single inventive concept. On the other hand, the USPTO tends to allow for broader claims and sheds less emphasis on specific technical applications.

Conclusion

With the new revisions in place, there are not yet enough case laws to exemplify CNIPA evaluation framework for AI inventions. Based on the examination guideline, China’s approach to patenting AI and algorithms might align more closely with Europe’s framework—we could lean more on the EP claims application to draft the Chinese version. They both emphasize the technical character and contributions of an invention, requiring that algorithms provide a technical solution to a specific problem. In writing Chinese patent applications, it might be prudent to lean closer to the EP patent application wording. In both applications, emphasizing technical specificity, clearly defining and centering the technical means and the specific technical problems the invention solves (i.e.: tangible technical effects and improvements like efficiency gains or novel applications in technical fields) could be beneficial. Along the same vein, it may also be advantageous to have a comprehensive disclosure of both technical and algorithmic features, using precise technical language. However, since inventiveness is not just defined by the technical character under CNIPA guidelines, more emphasis could be put on the inventiveness of the invention as a whole. Overall, following closer to the European framework could enhance the defensibility of the patent claim within the Chinese jurisdiction.

Navigating China’s legal landscape, particularly concerning patent law in the realm of AI and LLM, presents its unique challenges and complexities. However, it is also inevitable for companies to venture into China, given its massive market potential and rapidly growing technology sector. China’s commitment to fostering an innovation-driven economy, coupled with its robust capital markets and substantial investments in research and development, make it an incredibly attractive destination for companies seeking to protect and commercialize their AI-related intellectual property. Companies seeking IP protection must recognize the distinct nuances of Chinese patent law and the evolving nature of regulations surrounding emerging technologies and emerging governmental interest in this industry.

Given the intricacies involved, it is imperative for businesses to enlist the expertise of a legal team proficient in Chinese patent law or collaborate with local professionals who can guide them through the full process of patenting in China. Working with a legal team proficient in the Chinese legislature and ecosystem can help companies mitigate risks, optimize their patent strategies, and capitalize on the vast opportunities presented by China’s rapidly evolving AI ecosystem.

Thanks to intern Chloe Suyi Chan for her contribution to this article.

Scan the QR code to subscribe to "King & Wood Mallesons" for more information