Managing IP is part of the Delinian Group, Delinian Limited, 4 Bouverie Street, London, EC4Y 8AX, Registered in England & Wales, Company number 00954730
Copyright © Delinian Limited and its affiliated companies 2023

Accessibility | Terms of Use | Privacy Policy | Modern Slavery Statement

What are the Chinese examination standards for AI-related inventions?

Sponsored by


Quan Kang of Deqi details the examination standards for subject matter, inventiveness and sufficient disclosure in the arena of AI-related inventions and makes some suggestions for drafting claims and specifications

Artificial intelligence (AI) is based on data storage and machine algorithms. It is intended to simulate human thought processes and intelligent behaviours through using a computer. In recent years, with the rapid development of AI technology, the number of related patent applications has also increased correspondingly. In order to address issues arising from the progress of AI technology, the newly revised Chinese Guidelines for Patent Examination (hereinafter referred as Examination Guidelines), which were implemented on February 1 2020, clearly define the examination standards for AI-related inventions.

Types of AI-related inventions permitted in China

The Chinese National Intellectual Property Administration (CNIPA) deems that there are two types of AI-related inventions: inventions by means of AI and inventions created by AI independently. The first type can be divided into two sub-types: 1) inventions where the problem is raised and the scheme is worked out by a natural person, and the processes and effects are verified by AI; 2) inventions where the problem is raised by a natural person and other work is completed by AI independently.

The first type, i.e. inventions by means of AI, is permitted in China currently. Inventions created by AI independently are not allowed yet. There are no clear standards on how to examine the second kind of invention. The principle of the CNIPA on this point is essentially the same as that of the European Patent Office (EPO).

Examination standards relating to the subject matter of AI-related inventions

The examiner at the CNIPA first judges whether a claim as a whole of an AI-related invention is excluded from the range included in the "rules and methods for mental activities" under Article 25.1 of the Chinese Patent Law, and then examines whether it belongs to the technical solution category in Article 2.2 of the Chinese Patent Law.

Article 25.1 of the Chinese Patent Law

Article 25 provides a list of subject matters for which no patent right shall be granted.

Article 25.1: For any of the following, no patent right shall be granted:

……(2) rules and methods for mental activities; ……

Specifically, with regard to AI-related inventions, it is prescribed in Section 6.1.1, Chapter 9, Part II of the latest Examination Guidelines that, if a claim involves abstract algorithms or simple business rules and methods, and does not contain any technical features, the claim belongs to the rules and methods for mental activities in Article 25.1(2) of the Chinese Patent Law, and shall not be granted a patent right. If the claim contains a technical feature in addition to algorithm features or business rules and method features, then the claim viewed as a whole does not belong to the rules and methods for mental activities.

As long as there is a technical feature included in the claim, it shall not be excluded from patentability under Article 25.1(2).

Article 2.2 of the Chinese Patent Law

Article 2.2 provides for subject matters for which patent rights may be granted.

Article 2.2: "Invention" means any new technical solution relating to a product, a process or improvement thereof.

In particular, for AI-related inventions, according to 6.1.2, Section 6, Chapter 9, Part II of the latest Examination Guidelines, when examining whether a claim containing algorithm features is a technical solution, it is necessary to consider all the features recorded in the claims. If a step in the algorithmic feature of a claim relates closely to a technical problem to be solved, the execution of the algorithm could directly reflect the process of solving the technical problem in observance of the natural law and obtains technical effects, then this claim belongs to the category of a technical solution.

It is clear that for an AI-related invention, there are three elements used to decide whether it is a technical solution: the technical means involved in the claim, the technical problem solved and the technical effect obtained. For example, in a training method for a convolutional neural network (CNN) model, it is defined in the claim that the data processed in each step of the training method is image data. It also deals with how to process the data, which shows that the training algorithm relates closely to the image information processing. The problem to be solved is how to overcome the defect that the CNN model can only recognise images of a fixed size. Different processing and training methods are carried out here for images on different convolution layers. The result is that the trained CNN model in this claim can recognise images of any size. Therefore, this method belongs to the technical solution category and can be protected by a patent in China.

Examination standard for inventiveness of AI-related inventions

In Section 6.1.3, Chapter 9, Part II of the latest Examination Guidelines, it is stated that in the examination of inventiveness of AI-related inventions, the overall consideration principle and relevance consideration principle should be followed.

Overall consideration principle

All the features in the claim should be considered, including technical features, as well as algorithmic features or business rule and method features. Contribution of technical features and non-technical features should both be considered.

Relevance consideration principle

Technical features as well as algorithmic features or business rule and method features, supporting each other and having mutual relations should be considered as a whole. In the examination of inventiveness, the mutual support and interactions between technical features and non-technical features should also be considered.

A positive example is a claim which relates to detection technology, in which parameters and formulas of the algorithm are re-selected and adjusted. The application scenario of the claim is the same as that of the reference document, the difference between them only lies in the algorithm. The change of the algorithm enables the technical solution of the claim to improve the accuracy of detection. Thus, the claim possesses inventiveness in view of the reference document.

Another example is in the schemes of a claim and the reference document, the same algorithm is adopted, while the application scenarios are scenario A1 and scenario A2, respectively. In this circumstance, it is necessary to consider whether there are technical difficulties for those skilled in the art to obtain scenario A1 of the claim according to scenario A2 of the reference document, and whether there are technical effects obtained by applying the algorithm to scenario A1 etc. If it can be proved that the conversion from scenario A1 to scenario A2 is difficult and an unexpected technical effect is obtained, then the scheme of the claim has inventiveness.

The overall consideration principle on inventiveness of the CNIPA is essentially the same as that of the EPO.

Examination standard for sufficient disclosure of AI-related inventions

Article 26.3 of the Chinese Patent Law stipulates the requirement of clarity and completeness of the written description. The AI-related invention has unique characteristics in this respect.

Article 26.3: The description shall set forth the invention or utility model in a manner sufficiently clear and complete so as to enable a person skilled in the relevant field of technology to carry it out; where necessary, drawings are required.

"Black box" issue

The essential factors for sufficient disclosure in the above provisions involve "clarity" and "completeness". For AI-related inventions, the input initial data is clear, and the algorithm or model is also determined. However, sometimes it is difficult to know how AI is executed by the machine and what the logical relationship between the input data and the output data is, which may lead to the "black box" problem. If the specification does not clearly and completely disclose the concept of the creation, the essential technical features (algorithm, training data, each step of the computer program, etc.) related to the key invention points to make those skilled in the art able to program correspondingly and correctly, the application might be rejected by the CNIPA.


In the implementation of AI-related technical solutions a common situation occurs. There is the same input data, the same algorithm, the same machine training process/model, but the output results may not be the same. That is, the results cannot be reproduced.

In this case, the CNIPA aims to encourage inventive and creative behaviours. For patent applications that have sufficiently disclosed the key technical points of the invention, although the results may not be exactly the same, it is acceptable in China if the relevant data, algorithm and model have been fully described and defined. For example, if the core technology of the invention lies in machine learning and training, when all the training data and the machine learning process are fully recorded, even if the results are different within a certain range, the specification could be deemed as meeting the sufficient disclosure requirement and could be accepted by the Chinese examiner.

Suggestions for drafting claims and specifications

Categories of AI-related invention claims permitted by the CNIPA

Permitted: a method, an apparatus/a device, a system, and a computer storage medium, and so on. For example, a common method claim with actions and steps; a product claim including a program; a claim including program structures; a claim including "medium" and "computer program flow".

Not permitted: a computer program, a program product, a computer instruction.

Technical features in claims

According to Section 6.3, Chapter 9, Part II of the Examination Guidelines, for AI-related inventions, technical features and algorithmic features or business rule and method features that support each other and interact with the technical features in their function, must be clearly defined in the claims. Specifically, besides the technical features and algorithm features or business rule and method features, the interaction and interactive process of the features must be clearly defined in the claims.

In addition to the above requirements, claims of AI-related inventions currently must meet general requirements for other types of invention claims in various laws and regulations in China.

Contents in the description

The following contents should be recorded in the description:

1) The abstractive algorithm and its specific technique, in which the definition of at least one input parameter and its output results should be associated with specific data in the art.

2) Detailed technical schemes using natural language, with some code examples; when explaining the meaning of the code, natural language should not be replaced by the source code.

3) The effects in detail of the AI-related invention and the prior art; if there are unexpected effects, experiment data or simulation data should also be recorded.

4) As many technical details as possible of alternative schemes in the embodiment parts.

The CNIPA believes that with the development of AI technology, the person skilled in the art is required to have stronger capabilities. The scope of objects that can be patent protected has been expanded in Chinese patent law and the criteria for AI-related inventions has been loosened slightly.


Quan Kang

Quan Kang focuses on drafting and prosecuting patent applications, re-examinations, patent analysis, invalidations and litigation. She was named a Beijing Outstanding Patent Attorney in 2015, and was elected to be a teacher at the Capital Patent Agency Teacher Library.

more from across site and ros bottom lb

More from across our site

Cyril Amarchand Mangaldas has hired former Anand & Anand partner Swati Sharma and hopes to compete with specialist IP firms
Rapporteur-Judge András Kupecz ruled that education and training weren’t legitimate reasons for a member of the public to access documents
Searches for comparison prior art will be a little easier, but practitioners will have to put more thought into claim construction and design patent titles
The Helsinki local division rejected AIM Sport’s request for a preliminary injunction in a dispute with rival Supponor
We provide a rundown of Managing IP’s news and analysis coverage from the week, and review what’s been happening elsewhere in IP
The FTC’s plans to scrutinise improperly listed Orange Book patents could make these listings more important in litigation, but firms should be looking at this anyway
Counsel at Debevoise & Plimpton explain how they helped food delivery business Grubhub avoid a preliminary injunction at the Court of Appeals for the Seventh Circuit
European lawyers tell Managing IP how the legal market is reacting to the first few months of the UPC and why cases are set to take off
The ban could be extended or cancelled, depending on whether Judge Pauline Newman cooperates with an investigation, the Judicial Council of the Federal Circuit stated
Sources say some China-based lawyers are prepared to take large pay cuts to join stable practices, but most firms are sceptical about new hires