The rapid advancement of artificial intelligence (AI) and its broad integration across industries have contributed to a continued rise in AI-related patent filings worldwide. To enhance transparency and consistent examination practices, the Taiwan Intellectual Property Office (TIPO) released the Collection of AI-Related Invention Cases on October 13, 2025. Notably, the exemplary cases addressing subject-matter eligibility and sufficiency of disclosure have attracted particular attention from the IP community. This article provides an initial overview and introduction to the subject matter eligibility cases featured in the Case Collection.
In parallel, the Japan Patent Office (JPO) revised its Examination Handbook for Inventions and Utility Models on March 13, 2024, adding new AI-related exemplary cases concerning subject matter eligibility. Given that Taiwan's Software Examination Guidelines adopt the same analytical framework as Japan's, the updated JPO cases offer meaningful point of comparison. Therefore, this article also includes observations about the JPO's recent updates
► Two-Step Eligibility Test
TIPO's Collection of AI-Related Invention Cases includes two cases on subject-matter eligibility, analyzed under the following two-step procedure:
Step (1): Examiners should first determine whether the claimed invention clearly meets or clearly does not meet the aspects of the definition of invention. If a determination cannot be made, then the examination should proceed to Step 2.
Step (2): Determine whether the claim satisfies the requirement that “information processing via software is concretely realized through the utilization of hardware resources.”
TW Case 1– Calculating Mattress Match Score
In this exemplary case, Claim 3 recites an “information acquisition module,” an “AI matching module,” and a “display module,” and specifies how information is transferred among them. Since the claim outlines a technical means involving the collaborative operation of software and hardware, and describes a specific information processing method constructed based on the purpose of the information processing, it satisfies the Step (2) requirement of “information processing via software is concretely realized through the utilization of hardware resources.” Accordingly, Claim 3 meets subject-matter eligibility.
The Case Collection also clarifies that Step (2) does not require explicit recitation of hardware resources. It explains: If the claim recites a specific technical means of information processing, it is understood—based on common general knowledge at the time of filing—that such processing is implemented through the cooperative operation of general-purpose hardware and software. In this instance, the requirement of “information processing via software is concretely realized through the utilization of hardware resources” is satisfied. Conversely, Step (2) is not met if no specific information processing is described, even if hardware resources are expressly recited.
In essence, for software inventions involving the collaborative operation of general-purpose hardware and software, whether the claim recites “specific information processing” is the key factor in determining subject matter eligibility.
Claim 4 of TW Case 1 recites a sensor automatically acquiring a user's body type feature information, and the AI matching module receiving the body type feature information captured by the sensor to calculate a mattress match score for the user. Because “body-type features” constitute information relating to the technical properties of an object, the claim is deemed to perform information processing based on such properties and therefore satisfies Step (1).
Under Taiwan's Software Examination Guidelines, “technical properties of an object” refer to the physical, chemical, biological, electrical, or other such properties of an object. Thus, if the claimed technical invention involves information processing related to these technical properties, it can easily pass the examination for subject matter eligibility.
TW Case 2 – Predicting Program Memory Access Addresses
In this exemplary case, Claim 2 recites a data acquisition unit, a computing unit, and a recurrent neural network (RNN) unit implemented on an integrated circuit, which are configured to execute program instructions for steps such as acquiring memory access addresses, calculating address difference values, and generating feature vector representations for corresponding sequential data. The claim is considered to satisfy the Step (2) requirement because it involves the collaborative operation of software and hardware, and recites information processing, specifically “generating a feature vector... and predicting... based on... difference values...” Therefore, Claim 2 meets the definition of invention.
► JPO's New Subject Matter Eligibility Cases
Turning to the Japan Patent Office (JPO), the updated Examination Handbook for Inventions and Utility Models introduces new AI-related cases concerning subject matter eligibility, namely Case 5 and Case 2-14'.
JP Case 5 – Training Data and Method for Generating Training Data Images
Claim 1 of this case calls for a set of “training data” itself. Claim 2 calls for steps to obtain such data, such as acquiring a human body radiographic image and a surgical instrument image, performing image composition of these images by an image generation apparatus, and then generating images for use as training data.
Therefore, Claim 1 is deemed ineligible because it is merely a presentation of information. In contrast, Claim 2 is considered eligible because the “information processing by software is concretely realized by utilizing hardware resources.”
JP Case 2-14' – Training Model for Analyzing Accommodation Reputation
The newly added Case 2-14' serves as a counterpart to the existing Case 2-14. In the existing Case 2-14, Claim 1 was found eligible because it recited a specific calculation performed by the training model to analyze accommodation reputation, thereby utilizing hardware resources for specific realization.
In comparison, the new Case 2-14' is deemed ineligible because Claim 1 recites that “the model is composed of a set of parameters” and “a set of parameters” is classified as a presentation of information and does not constitute a program or software capable of producing specific technical results. Accordingly, it fails to meet the subject-matter eligibility requirement.
► Key Takeaways from TIPO and JPO Updates
The newly introduced AI-related cases highlight a key distinction between the two offices:
- The JPO cases provides exemplary scenarios where applications fail to meet the Step (1) subject-matter eligibility requirement, clarifying that “training data itself” and a “set of parameters” are merely presentations of information rather than “creations of technical ideas utilizing a law of nature.”
- The TIPO cases place greater emphasis on Step (2), reinforcing that claims must recite a concrete information-processing method involving the cooperative operation of software and hardware.
Overall, the updates offer practical guidance for both examination and application practices with respect to the eligibility test.