China’s National Intellectual Property Administration (CNIPA) issued a decision amending the Patent Examination Guidelines (Order No. 84). One core change further clarifies the examination rules for invention applications involving AI, big data and similar subject matter containing algorithm features or business-rule and method features.
Three changes worth noting
First, the section title was updated. The amendment retitles section 6 of Chapter 9, Part II to “examination provisions for invention applications involving algorithm features or business-rule and method features, such as artificial intelligence and big data”, writing “AI and big data” explicitly into the title of the examination rules. That is itself an institutional signal: AI-related applications are now an independent category in patent examination.
Second, the examination baseline was refined. After the amendment to section 6.1, examination should be directed at the solution defined by the claims, and “where necessary, the description should also be examined”. At the same time, examiners should not simply sever technical features from algorithm features and business-rule and method features. This means the algorithm and the technical solution are to be viewed as a whole, rather than rejecting technical character merely because an algorithm is involved.
Third, a new exclusion was added. Section 6.2 adds that invention applications containing algorithm features or business-rule and method features that violate law, social morality or harm the public interest cannot be granted. This draws a bottom line for the grant of AI-related applications.
What this means for AI patent strategy
First, the description must show the solution as a whole. Since the algorithm is no longer viewed in isolation, the description should explain how algorithm features combine with technical means, what technical problem is solved and what technical effect is achieved. Describing the algorithm alone without technical problem and effect may still fail the subject-matter test.
Second, watch the risk created by the exclusion. The new clause makes “contrary to law, social morality or public interest” a ground for exclusion. For applications in face generation, deepfakes or algorithmic discrimination, the description should avoid falling within that scope, and product positioning and patent drafting should be consistent.
Third, combine patents with trade secrets. For core algorithms that are hard to disclose sufficiently while preserving technical effect, trade secret protection may be considered, with the patent portfolio focused on disclosable technical solutions and system architecture. See our IP protection service.
Echoing the copyright debate
Beyond patents, AI copyright issues are advancing in parallel. Shanghai courts’ AI prompt copyright case and China’s first AI model infringement case are accumulating judicial experience on ownership and infringement boundaries for generative AI. See our notes on the AI prompt copyright case and the first AI model infringement case, and on AI copyright criminal liability.
What companies should do
For AI companies, the amendment brings clearer expectations: where algorithm features form a whole with the technical solution and solve a technical problem, there is room for grant; but where the legal and public-interest bottom line is crossed, exclusion is explicit.
Before filing, companies should map the “algorithm — technical problem — technical effect” relationship in the solution, assess whether the application direction touches the exclusion, and arrange patents and trade secrets together for core algorithms.
This article is based on CNIPA’s decision amending the Patent Examination Guidelines (Order No. 84) and public commentary. It is for general reference only and does not constitute legal advice.