Intellectual Property 2026

JAPAN Law and Practice Contributed by: Takahiro Inoue, Tomohiro Kuribayashi, Yoshiyuki Takanashi and Daisuke Inaba, TMI Associates

7.3 Training Data, Model Development and Infringement Risk The use of copyrighted works for AI training involves reproduction and may therefore constitute infringe- ment unless it falls within a copyright exception. In this context, Article 30-4 of the Copyright Act provides that the use of works for the purpose of information analysis (including AI training) is permissible if the use: • is not for the purpose of enjoying the thoughts or sentiments expressed in the work; • is limited to the extent necessary for such purpose; and • does not unreasonably prejudice the interests of the copyright holder, having regard to the nature or purpose of the work and the circumstances of its use. In general, the use of copyrighted works for AI training is permissible under Article 30-4, unless the above requirements are not satisfied. Although no court rul- ings exist on Article 30-4, the AI-Copyright Guide- lines suggest, for example, that AI training aimed at generating outputs that reproduce all or part of the creative expression in the training data fails the first requirement above and may therefore constitute infringement. They further indicate that, where a cop- yrighted database that contains data organised in a format suitable for information analysis and is made available on the internet for a fee is reproduced for AI training without compensation, such conduct may fail the third requirement and may therefore constitute infringement. Furthermore, where trade secrets protected under the UCPA are used for AI training and such trade secrets are divulged in AI-generated outputs, this may con- stitute misappropriation of trade secrets in violation of the UCPA. There is currently no law or regulation that imposes a legal obligation to disclose information regarding the data used in training datasets. 7.4 Enforcement Against AI-Enabled Infringement There is no specific legal mechanism addressing infringements involving generative AI, and the absence

The scraping or data extraction of copyrighted works may constitute copyright infringement. However, scraping for certain limited purposes, such as data analysis, may be considered permissible under appli- cable copyright exceptions (Articles 30-4 and 47-5). Furthermore, data and databases may be protected as “trade secrets” under the UCPA, and the acquisition, use or disclosure of such data or databases through wrongful means may constitute unfair competition. In addition, the relevant parties may contractually pro- hibit the disclosure of such data or databases to third parties or their use for specified purposes. 7.2 AI-Generated Works and Inventorship/ Authorship Copyright Although no judicial precedent exists on the copy- rightability of AI-generated outputs, the Legal Sub- committee under the Copyright Subdivision of the Cultural Council (a governmental advisory body under the Agency for Cultural Affairs) published guidelines entitled “General Understanding on AI and Copyright” (the “AI-Copyright Guidelines”), on 15 March 2024. While not legally binding, the AI-Copyright Guidelines serve as an important practical reference in Japan, setting out a framework for issues such as whether AI- generated outputs are copyrightable. According to the AI-Copyright Guidelines, materials autonomously gen- erated by AI are not copyrightable. However, where a person creates a work using AI as a tool and makes a creative contribution, considering factors such as the amount of instructions or inputs, the number of generation attempts, and selection from outputs, such a work may be copyrightable. Where copyrightability is recognised, the AI user who created the output is considered as the author and holds both copyright and moral rights. Patent AI-related technologies may be patentable, but inven- tions autonomously generated by AI are not. In a case where an engineer named an AI as the inventor, the Intellectual Property High Court, in a decision dated 30 January 2025, held that an “inventor” must be a natural person, which was subsequently upheld by the Supreme Court.

169 CHAMBERS.COM

Powered by