AI has become part of the process of invention-creating and patent-drafting activities. The regulatory provisions are mainly set out in the Guidelines for Filing AI-Related Invention Patent Applications (for Trial Implementation) (the Guidelines), issued by the CNIPA in December 2024, and the Opinions on Lawfully Adjudicating Cases Involving Artificial Intelligence Disputes (Fa Fa (2026) No. 10) (the Opinions), issued by the Supreme People’s Court of the People’s Republic of China in September 2026.
General principle: use of AI is permitted
China’s current legal framework does not prohibit the use of AI tools in the conception of inventions or the drafting of patent applications. In Chapter One of the Guidelines, the CNIPA explicitly recognises “AI-assisted inventions” as a distinct category of patent applications. Such inventions are defined as those “achieved through the use of AI technology as an auxiliary tool during the inventive process”, wherein AI functions analogously to an information-processing or drafting aid rather than as an autonomous inventor.
With respect to patent drafting support, an official response published on CNIPA’s website on June 23 2026 expressly permits specific auxiliary activities, including prior art searching, linguistic refinement of application texts, and formatting adjustments.
Core risks
The inventor must be a natural person
AI technology can be used as an auxiliary tool during the inventive process, but it cannot be listed as an inventor. The heading of Chapter Two of the Guidelines is “The Inventor Named in the Application Must Be a Natural Person”. The second sentence of Article 14 of the Opinions takes the same position. It follows that a patent application naming AI as the inventor will be rejected; in Decision on Request for Re-examination No. 1373038 of the CNIPA (the DABUS case), the rejection was upheld.
An invention generated by AI to which no natural person has made a substantive contribution cannot be granted a patent (Article 14 of the Opinions). In other words, where no natural person has made a creative contribution to the substantive features of an invention-creation, that invention-creation can neither be attributed to a person as its inventor nor be granted a patent. If the R&D process is fabricated to satisfy these requirements, the principle of good faith laid down in Article 20 of the Patent Law is violated.
Risks of information leakage, substantive defects, and findings of bad faith
On April 1 2026, the CNIPA issued three risk warnings regarding the use of AI in drafting patent application documents.
First is the risk of technical information leakage, wherein invention disclosure, experimental data, and technical drawings contain undisclosed information and uploading them to a publicly accessible AI platform creates a risk. For the applicant, premature disclosure compromises the novelty requirement under patent law, thereby rendering the invention unpatentable. Moreover, third parties may exploit the disclosed information to file competing applications, potentially undermining the applicant’s rights.
Second is the “substantial defect” risk, wherein AI-assisted drafting may introduce internal logical inconsistencies or inadequately describe technical features, thereby undermining the clarity, support, and enablement requirements essential for obtaining valid patent protection.
Third is the risk of a finding of bad faith: “patent applications generated entirely by AI through fabrication, random making-up, or content patchwork constitute dishonest and unethical behaviour that violates the principle of good faith”. Such applications will not be granted; applicants may face warnings or fines, while patent agencies and patent attorneys engaging in such misconduct could have their practice licences revoked or agency qualifications cancelled.
Risks of copyright infringement
Since specifications and drawings may constitute works when meeting certain conditions, care must be taken to avoid incorporating others’ prior expressions when using AI in generating content, thereby preventing copyright infringement risks. Article 12 of the Opinions states that users who “know or should know about the existence of prior works and use AI to generate substantially similar works without a valid defence” shall bear liability for infringement. Chapter 6 of the Guidelines also requires attention to be paid to the sources and usage standards of relevant data.
Conclusion and recommendations
AI has practical value in invention-creating and patent-drafting activities, but its use requires practitioners to assume a heavier duty of review and oversight. The following measures are therefore recommended:
Mandatory human review should be put in place. AI-generated content must be verified item by item by the attorney before it is filed.
Records of human contribution should be kept, so as to be able to answer any challenge to inventor qualification.
Clear rules on use should be adopted at the institutional level and input strictly controlled. Invention disclosures, experimental data, and drawings must not be entered into a publicly accessible AI platform. If processing is necessary, tools deployed locally or professional tools capable of providing binding confidentiality commitments should be used.
