LEGAL UPDATE | NO. 3 | SEPTEMBER 2026
[SAKURA Law Office | Legal Update by Managing Partner Kenshiro Michishita]
SAKURA Law Office
Managing Partner Kenshiro Michishita
Tokyo, Japan | September 24, 2026
Executive Summary
Copyright issues involving generative AI in Japan must be analyzed by separating at least three questions: the use of copyrighted works at the AI development and training stage; infringement risk when AI-generated outputs are created and commercially used; and whether the AI-generated output itself qualifies for copyright protection. Article 30-4 of the Japanese Copyright Act may permit certain uses of copyrighted works for information analysis and other non-enjoyment purposes, but it is not a blanket exemption for all AI training. At the output stage, ordinary copyright principles — including similarity and reliance on an existing work — remain central. For businesses, the practical task is therefore not merely to ask whether “AI use is permitted,” but to design a repeatable process covering training data, RAG, input materials, output review, service terms, ownership, contractor arrangements and evidence preservation.
SAKURA Law Office has published the third installment of the Legal Update series by Managing Partner Kenshiro Michishita, entitled “Generative AI and Copyright in Japan — Key Legal Issues in AI Training, AI-Generated Content and Commercial Use (2026).”
As corporate use of generative AI becomes routine, the practical questions have become increasingly specific: May a company use copyrighted works as AI inputs? May AI-generated images or text be used in advertisements or products? Does copyright arise in an AI-generated output? Is a company safe merely because an AI provider’s terms state that commercial use is permitted?
One reason these questions are difficult is that Japanese copyright law applies differently at the development and training stage than at the generation and use stage. A separate question then arises as to whether the AI-generated output itself qualifies as a copyrighted work and, if so, who should be regarded as the author or copyright owner.
In March 2024, Japan’s Agency for Cultural Affairs published the “General Understanding on AI and Copyright in Japan,” followed by checklists, guidance and other explanatory materials. The General Understanding is not legally binding, and Japan still has limited domestic case law addressing copyright questions unique to generative AI. Corporate decisions must therefore be grounded in the Copyright Act and established copyright principles, while also taking account of the technical design of the relevant AI service, the generation process and the intended use of the output.
The policy environment has also continued to develop. In June 2026, the Intellectual Property Strategic Program 2026 was adopted. On August 25, 2026, the “Principle Code on the Protection of Intellectual Property and Transparency for the Appropriate Use of Generative AI” was published. These developments show that the relationship between AI and intellectual property is increasingly being addressed through a combination of law, technology, contract, transparency, remuneration mechanisms and governance.
This Legal Update explains the principal copyright issues that companies should consider when developing, deploying or using generative AI in Japan as of September 24, 2026, following the sequence in which those issues typically arise in practice.
1. Separate the Development and Training Stage from the Generation and Use Stage
Copyright analysis should begin by identifying the stage of the AI lifecycle in which the relevant conduct occurs.
The first stage is the development and training stage. This may involve collecting, reproducing and analyzing books, articles, images, videos, music, software and other copyrighted works to train or fine-tune an AI model. At this stage, the principal question is often whether Article 30-4 of the Copyright Act or another statutory limitation on copyright applies.
The second stage is the generation and use stage. Here, a user enters prompts, images, documents or other information into a trained model and produces outputs that may then be saved, reproduced, internally distributed, sold, published, publicly transmitted or otherwise exploited. At this stage, the central issues include whether the output is sufficiently similar to an existing copyrighted work and whether it was created in reliance on that work, as well as whether any exclusive rights such as reproduction, adaptation or public transmission rights are infringed.
A third and distinct issue is whether the AI-generated output itself qualifies as a “work” protected by copyright and, if so, who may be regarded as the author. That is a different legal question from whether the output infringes someone else’s copyright.
Keeping these three questions separate is the starting point for reliable corporate analysis.
2. Article 30-4 of the Copyright Act Is Not an Unlimited AI-Training Exception
Article 30-4 of the Japanese Copyright Act permits the exploitation of a copyrighted work, within the scope considered necessary, where the purpose is not to personally enjoy or cause another person to enjoy the thoughts or sentiments expressed in the work. Data analysis is one of the examples expressly contemplated by the statute.
Because machine learning may involve reproducing and analyzing works for information-analysis purposes rather than for appreciation of their expressive content, Article 30-4 may apply to certain AI-training activities.
However, Article 30-4 is not a blanket rule that makes every use of a copyrighted work for AI training lawful. If the use also has a purpose of enjoying the expressive content of the work, the premise of the exception may not be satisfied. Even where the use is non-enjoyment in nature, the proviso to Article 30-4 excludes use that would unreasonably prejudice the interests of the copyright owner in light of the nature or purpose of the work and the circumstances of its exploitation.
Accordingly, a company developing or fine-tuning an AI model should not conclude that “AI training is automatically lawful under Article 30-4.” It should examine the specific purpose of the training, the datasets involved, the nature of anticipated outputs, the existence of licensing markets and other circumstances relevant to the statutory analysis.
3. Training Designed to Reproduce the Creative Expression of Particular Works Requires Greater Caution
Even where an AI-training activity has an information-analysis purpose, a purpose of enjoying the creative expression of copyrighted works may coexist with that analytical purpose.
For example, where a dataset of particular works is assembled specifically so that the system can generate outputs reproducing the creative expression of those works, the analysis may differ materially from ordinary statistical or analytical use.
The proviso to Article 30-4 may also become relevant where an existing or reasonably anticipated licensing market would be materially displaced. The Agency for Cultural Affairs has discussed, among other examples, the reproduction of database works for AI training in circumstances where licenses for data analysis are commercially available.
Companies using third-party content, professional databases, article archives, stock images, source code or other large collections for training should therefore review not only statutory copyright exceptions but also terms of service, database licenses, API terms, technological access controls and contractual restrictions.
4. RAG and Summarization Services Are Not Automatically Covered by Article 30-4
Retrieval-augmented generation, commonly referred to as RAG, is increasingly used to allow AI systems to search internal documents, books, articles and other source materials before generating an answer.
The Agency for Cultural Affairs has indicated that, where copyrighted works are collected as input data for a RAG system and the system is designed so that the creative expression of the works is not reproduced in the output, Article 30-4 may apply to the collection and reproduction involved.
By contrast, where the system is designed to generate and present the creative expression of the underlying works to users, a purpose of enabling users to enjoy that expression is involved, and Article 30-4 may not apply.
It is therefore unsafe to assume that an internal RAG system or a summarization function is lawful merely because it is described as search or summarization. Companies should examine how source materials are copied and stored, how much of the original expression may appear in outputs, who will receive those outputs, and whether the relevant materials are subject to licensing restrictions.
Particular care is required before ingesting paid articles, subscription databases, contract templates, research reports, commercially published books or other licensed materials into a RAG environment.
5. Companies Should Avoid Using Pirated or Otherwise Infringing Content as Training Data
The provenance of training data matters. Content available on the internet is not necessarily content that has been lawfully uploaded or made available.
The Agency for Cultural Affairs has noted that the application of Article 30-4 to infringing reproductions, including pirated copies, requires assessment in the circumstances of the individual case. At the same time, knowing use of pirated materials as training data may become relevant where an AI system later produces infringing outputs and may increase the risk that an AI developer is alleged to bear responsibility for infringement.
As a matter of corporate practice, there is rarely a sound business reason to build a training pipeline around sources with serious provenance concerns. Companies should, to the extent reasonably possible, document the origin, acquisition method, licensing status and rights position of training data and should avoid piracy sites, unlawful uploads and other sources where the legitimacy of the content is materially doubtful.
Japanese intellectual-property policy in 2026 also places increasing emphasis on lawful datasets, transparency and appropriate remuneration or value return to rights holders. The ability to explain data provenance is likely to become increasingly important to commercial credibility, contracting and enterprise procurement.
6. At the Output Stage, Copyright Infringement Continues to Turn Principally on Similarity and Reliance
When a company uses AI-generated text, images, music, video, software or other outputs, infringement is generally assessed under ordinary copyright principles. Two central questions are similarity and reliance.
Similarity concerns whether the essential characteristics of the creative expression of an existing work can be directly perceived in the later output. Similarity in ideas, facts, commonplace expression or other unprotected elements does not by itself establish copyright infringement.
Reliance concerns whether the later expression was created on the basis of the pre-existing work. Where the AI user knew of an existing work and intentionally referred to it in order to generate a similar output, reliance is more readily established.
There is no rule under Japanese law that an output is insulated from infringement merely because it was generated by AI. The final output remains subject to ordinary copyright analysis.
7. Image-to-Image and Other Direct Inputs Make Reliance Easier to Establish
Companies increasingly provide an existing image, text, recording or other copyrighted material directly to an AI system and ask it to transform, restyle or build upon that material.
In such cases, the user is plainly aware of the source work and has chosen to provide it as an input. If reliance becomes disputed, that factual link is therefore easier to establish than in a case where the user never encountered the source work.
Direct input does not mean that infringement automatically occurs. The analysis still requires consideration of similarity in protected expression, any applicable license and any statutory exception.
However, in advertising, product design, film production and other commercial workflows, a company should not assume that the phrase “used only as a reference” is legally sufficient. Input materials, prompts, generation histories and the final work should be documented so that the rights basis for the output can later be explained.
8. Reliance Is More Difficult Where the User Did Not Know the Source Work — but the Issue Remains Unsettled
A particularly difficult generative-AI issue arises where the user was unaware of an existing work, yet the AI system produces an output highly similar to it.
The Agency for Cultural Affairs’ General Understanding discusses reliance in circumstances where the user did not know the work but the work was included in the model’s training data. The analysis reflects factual circumstances that differ from many traditional copyright cases.
However, the General Understanding is an administrative interpretation rather than binding law, and Japan still has limited domestic case law addressing AI-specific reliance. In a particular dispute, factors may include whether the model had access to the work through training data, the degree of similarity, the nature of the model, the prompt, the generation history and other surrounding circumstances.
For important commercial outputs, companies should not rely on the hope that the user can later say that the source work was unknown. A more defensible practice is to perform similarity review, stop use where a highly similar existing work is identified, and consider regeneration or licensing where appropriate.
9. A Prompt Such as “In the Style of X” Is Not Automatically Copyright Infringement — but It Is Not Risk-Free
Generative AI is frequently instructed to produce content reminiscent of a particular artist, creator, film, animation style, brand or aesthetic.
Copyright law does not generally grant an exclusive right over abstract ideas, artistic style or visual manner as such. Similarity in mood or style alone therefore does not automatically establish copyright infringement.
However, if the resulting output reproduces concrete creative expression from a particular work — such as a specific composition, character expression, distinctive arrangement or protectable text — ordinary similarity and reliance analysis applies.
Other legal rights may also arise even where copyright does not protect the element in question. Depending on the facts, trademark law, the Unfair Competition Prevention Act, publicity rights, reputation, contract or other legal doctrines may be relevant. A voice as such is not ordinarily protected directly by copyright, but use of a particular performance, accompanying images or music, or a celebrity’s commercial identity may raise separate issues.
Where a company intentionally uses AI to approach the identity, brand value or creative signature of another creator or public figure in commercial advertising or products, a broader intellectual-property and reputational review is prudent.
10. “Commercial Use Permitted” in an AI Provider’s Terms Does Not Mean “No Third-Party Copyright Risk”
Many AI services state in their terms that users may commercially exploit generated outputs. That contractual permission should not be confused with a warranty that the output cannot infringe a third party’s copyright.
Even if the provider grants the user contractual rights to use an output, a separate copyright issue arises if that output was created in reliance on, and is sufficiently similar to, a third party’s existing work.
Some providers offer intellectual-property indemnities or other protection under certain enterprise plans. Those programs typically contain conditions concerning the eligible subscription, permitted uses, excluded conduct, claim procedures, liability caps and user obligations. The existence of an indemnity should not be treated as a substitute for rights review.
For material commercial uses, companies should review the provider’s allocation of rights in outputs, infringement warranties and indemnities, user-side indemnification obligations, prohibited prompts or inputs, and cooperation duties in the event of a claim.
11. Copyright in the AI-Generated Output Depends on Human Creative Contribution
Whether an AI-generated output infringes someone else’s copyright is separate from whether the output itself qualifies for copyright protection.
Under the Japanese Copyright Act, a copyrighted work must be a creative expression of human thoughts or sentiments. Where an AI system produces an output autonomously with little or no human creative involvement, the question arises whether that output can be treated as a human-authored work at all.
The Agency for Cultural Affairs has indicated that an output may qualify as a work of a human author where the AI is properly regarded as a tool used in human creation. The assessment is expected to consider whether the person had a creative intent and whether the person made a creative contribution throughout the generation process.
The legal significance of prompts depends on their substance. A short generic instruction is different from a detailed direction that concretely shapes expressive choices. Likewise, generating a large number of outputs does not, by itself, establish creative contribution, and simply selecting one result from many does not automatically create copyrightability.
Where a person makes creative additions, revisions, editing or composition after generation, the human-created portions may in appropriate circumstances qualify for copyright protection.
12. Companies Must Separately Design How AI-Generated Content Will Become Protectable Corporate IP
The rights analysis should differ depending on whether AI-generated content is a disposable internal draft or a long-term corporate asset such as a logo, character, product design, film, game, publication, software asset or key advertising creative.
If an output lacks sufficient human creative contribution to qualify for copyright, the company may have difficulty preventing competitors or third parties from using the same or similar content through copyright law alone.
Where an employee creates content with AI, even if human authorship is sufficient for copyright to arise, the company does not automatically become the author or copyright owner in every case. The requirements for corporate authorship under Article 15 of the Copyright Act, the nature of the employee’s duties, the name under which the work is made public, and relevant employment rules or policies may need to be considered.
The same issue arises when AI-assisted creative work is outsourced to an agency, designer, writer, film producer or other contractor. Paying a production fee does not, by itself, transfer copyright. Agreements should address ownership, assignment or licensing, non-exercise of moral rights where appropriate, disclosure of AI use, and the use of third-party materials.
13. Preserve Prompts, Input Materials, Generation History and Human Editing Records
In AI copyright disputes, the process by which the final output was produced can become as important as the output itself.
If a third party alleges infringement, relevant facts may include whether a source work was used as a prompt or reference image, which model and version were used, what instructions were given, how the output evolved, and how much human editing occurred. These facts may affect reliance, responsibility, copyrightability and contractual allocation of liability.
The same records may assist a company seeking to establish its own copyright. Evidence of the creator’s intent and the specific human decisions that shaped the expression may support an argument that the AI functioned as a creative tool rather than an autonomous author.
For important commercial AI-generated assets, companies should consider retaining, to a reasonable extent, the generation date, service and model, principal prompts, reference materials, generation history, reasons for selection, human edits and final approver.
14. Infringement Liability May Extend Beyond the User Depending on the Developer’s or Provider’s Conduct
Where an AI-generated output infringes copyright, the conduct of the person or company that generated, published, sold or otherwise exploited the output is an obvious starting point for liability analysis.
However, an AI developer or service provider is not necessarily irrelevant. Depending on the design of the service and the degree of involvement, liability may be argued where the system is intentionally designed to generate outputs similar to particular copyrighted works, where infringing generation is actively induced, or where the provider has concrete knowledge of infringement and fails to take appropriate measures.
Conversely, the mere fact that a provider offers a general-purpose AI service does not make it automatically liable for every infringing act of every user. The analysis may turn on each party’s conduct, control, knowledge, technical measures, contractual role and other specific facts.
For enterprise adoption, contractual allocation of third-party intellectual-property risk remains an important part of procurement and implementation.
15. In 2026, Japanese Policy Has Moved Further Toward Transparency and Rights Protection
The Intellectual Property Strategic Program 2026, adopted on June 12, 2026, reaffirmed Japan’s policy objective of promoting generative-AI innovation while appropriately protecting intellectual property rights.
The Program emphasizes that AI and intellectual property should not be addressed solely through legislation. It calls for an appropriate combination of legal, technological and contractual measures and cooperation among AI developers, providers, users and rights holders.
On August 25, 2026, the Cabinet Office Intellectual Property Strategy Promotion Secretariat published the “Principle Code on the Protection of Intellectual Property and Transparency for the Appropriate Use of Generative AI.”
The Principle Code is not legally binding. It operates on a comply-or-explain basis and is intended as a governance framework concerning transparency and intellectual-property protection for generative-AI developers and providers, including potentially overseas businesses offering generative-AI systems or services for Japan.
The Code addresses matters such as disclosure of an overview of AI systems and services, measures to protect intellectual property, and frameworks for responding in certain circumstances to requests from rights holders or users concerning whether identified URLs were used as training data. Notification to the Cabinet Office under the framework is scheduled to begin on October 26, 2026.
The Code does not itself determine whether copyright infringement has occurred. Its practical significance lies in increasing transparency and the possibility for rights holders and users to obtain information relevant to legal assessment.
16. Commercial Review Should Be Proportionate to the Importance and Exposure of the Output
Companies do not need to conduct the same level of copyright review for every AI-generated draft. The appropriate level of review should reflect the intended use and the consequences of a potential infringement.
An internal brainstorming note is materially different from a corporate logo, product package, advertising campaign, television commercial, film, song, character, game, publication or website hero image that will be distributed widely and used for an extended period.
For important external uses, companies should consider combining similarity review, verification of input rights, meaningful human editing, review of provider terms, preservation of the generation process and, where appropriate, specialist legal or IP review.
Outputs intentionally designed to evoke existing brands, well-known characters, celebrities or identifiable creators should generally be treated as higher-risk uses.
17. Contracts with External Creators and Production Companies Should Address AI Use Expressly
Advertising agencies, film producers, web designers, publishers, game developers and other contractors increasingly use generative AI in creative workflows.
If a client does not know that AI was used and a third party later alleges infringement, it may become difficult to determine what source materials were used, which AI system generated the content and who must bear responsibility.
For material creative engagements, agreements may therefore address prior approval or notification of generative-AI use, permitted AI services, prohibitions on unauthorized third-party works as inputs, lawful provenance of training or reference materials, obligations to secure necessary rights, representations concerning third-party rights, cooperation in disputes, indemnification and preservation of generation records.
A blanket prohibition on every use of AI may, however, become disconnected from real-world production processes and may be ignored in practice. Contract design should distinguish prohibited uses from controlled uses that are permitted under specified conditions, taking account of the value of the brand and the intended use of the content.
18. Rights Holders Should Think About Evidence Preservation as Soon as AI Infringement Is Suspected
Where a rights holder discovers AI-generated content that closely resembles its image, text, video, music, software or other copyrighted work, early evidence preservation is important.
The rights holder should preserve the disputed output, URL, publication date, publisher or account information, surrounding display, and any public information regarding the AI service or prompt used. It should then identify whether the shared elements involve protectable creative expression rather than merely ideas, facts or commonplace elements.
Where infringement is established, potential remedies may include injunctive relief, damages and other civil remedies. Depending on the facts, takedown requests to a platform or AI provider, complaints under terms of service and measures to identify the relevant poster or user may also be available.
For participating businesses, the transparency mechanisms contemplated by the 2026 Principle Code may in some circumstances provide an additional route for fact-finding regarding training data.
AI-related copyright disputes often require both technical and legal analysis. Early cooperation with technical specialists can be important when preserving evidence and evaluating how the disputed output was produced.
19. Frequently Asked Questions
Q1. Can a company freely use an AI-generated image for commercial purposes?
Not necessarily. The company should confirm both that the AI provider’s terms permit the intended commercial use and that the output does not infringe a third party’s existing copyright through prohibited similarity and reliance. If the output lacks sufficient human creative contribution, the company may also be unable to obtain strong copyright-based exclusivity over the output itself.
Q2. Is using online images for AI training always lawful under Article 30-4?
No. Article 30-4 permits certain uses where there is no purpose of enjoying the thoughts or sentiments expressed in the work, subject to statutory conditions. It does not apply where an enjoyment purpose is present or where the use would unreasonably prejudice the copyright owner’s interests. The purpose of the training, the nature of the data and the effect on existing markets should be assessed.
Q3. Is prompting an AI to create something “in the style of [a named artist]” copyright infringement?
Not automatically. Abstract style or artistic manner is not generally protected as such by copyright. However, infringement may arise if the output reproduces concrete protectable expression from a specific existing work. Other legal rights may also be relevant.
Q4. Does a small amount of human editing automatically make an AI output copyrightable?
No. The human changes must themselves amount to creative expression. Where creative additions or revisions are made, copyright may arise at least in those human-created elements, but the assessment is fact-specific.
Q5. If an employee creates an AI image for work, does the copyright automatically belong to the company?
Not necessarily. The first question is whether the output is copyrightable at all through sufficient human creative contribution. If it is, the requirements of Article 15 on corporate authorship, the employee’s duties, publication under the company’s name, internal rules and other circumstances should be examined.
Q6. If the AI provider offers an IP indemnity, can the company skip internal review?
No. Indemnities usually apply only to defined plans, permitted uses and specified claim conditions, and may contain exclusions and caps. They also do not necessarily address business interruption, replacement costs or reputational harm caused by an injunction or dispute. Material commercial uses still require appropriate rights review.
Q7. May we ingest books or paid articles that the company has purchased into an internal RAG system?
Purchase alone does not necessarily authorize reproduction, indexing, storage and generation through a RAG system. The company should review Article 30-4, whether the output reproduces protected expression, and contractual or licensing restrictions on copying, machine use or database ingestion.
20. The SAKURA Law Office Perspective on Generative AI and Copyright
Corporate copyright practice in the generative-AI era cannot be reduced to abstract questions such as whether “AI training is lawful” or whether “AI outputs are copyrightable.”
The practical questions are operational: which works will be used, for what purpose, in which AI system, under what license; whether the output will be commercially exploited; how similarity to existing works will be reviewed; whether the company can secure the rights it needs in the output; and who bears responsibility if a third party brings an infringement claim.
As generative AI reduces production costs and accelerates content creation, a single rights-clearance error can also propagate across large volumes of advertising, products and digital content. Copyright governance must therefore become as scalable and repeatable as the AI-enabled production process itself.
Companies should also remember that copyright management is not only about avoiding infringement. For logos, designs, text, video and other AI-assisted assets intended to create competitive value, the ability to secure and enforce rights may be equally important.
For material AI-generated assets, it is therefore prudent to manage infringement risk, human creative contribution, contractual ownership, related trademark and unfair-competition issues, and generation records as one integrated intellectual-property governance process.
21. About the Kenshiro Michishita Legal Update Series
SAKURA Law Office continues to publish the Legal Update series on the profile page of Managing Partner Kenshiro Michishita, addressing legal issues of practical importance to companies, management teams and business leaders.
The first Legal Update addressed the broader legal risks of corporate generative-AI use and the design of AI governance in Japan. The second installment addressed whether companies may enter customer or personal information into ChatGPT and other generative-AI services under Japan’s APPI.
This third installment focuses on copyright, an issue of comparable importance to data protection in enterprise AI deployment, and addresses AI training, RAG, commercial use of generated outputs, similarity and reliance, copyrightability of AI-generated content, contracting and Japan’s 2026 policy developments.
Future Legal Updates will continue to address confidential information and trade secrets, AI service agreements, internal AI policies, AI agents, M&A, cross-border transactions, Web3 and digital assets, crisis management, corporate misconduct, information governance and other areas of corporate law.
22. How SAKURA Law Office Can Assist with Generative AI and Copyright
SAKURA Law Office advises companies on copyright and other intellectual-property issues arising from the development, deployment and use of generative AI.
Our work may include reviewing whether AI-generated images, text, video, music or software may be commercially used; assessing third-party rights risk in AI-assisted advertising, products and content; reviewing the ingestion of books, articles, contracts and other materials into internal RAG or AI-search systems; structuring rights in training and fine-tuning datasets; reviewing intellectual-property and indemnity provisions in AI-service agreements; drafting AI-use clauses for production companies and external creators; structuring ownership and protection of AI-generated assets; responding to allegations of AI-related infringement; and enforcing rights where a company’s own works appear to have been infringed through AI-generated content.
Companies may consult us before the final AI service, dataset or production workflow has been selected. By reviewing the proposed use case, source materials, AI service, intended output, distribution model and contractual relationships, we can identify the rights-clearance, contracting, governance and dispute-prevention measures that should be addressed before deployment.
Generative AI / Copyright Legal Inquiries
For advice concerning AI training data, commercial use of AI-generated outputs, copyright, RAG, AI-assisted content production, AI service agreements, intellectual-property infringement, internal AI policies or other AI / IT legal matters, please contact SAKURA Law Office.
Corporate clients may consult us not only after a dispute has arisen, but also regarding pre-deployment legal reviews, output-clearance processes, production and outsourcing agreements, internal guidelines and ongoing AI / intellectual-property governance.
SAKURA Law Office
Managing Partner: Kenshiro Michishita
4F, Ark Hills South Tower, 1-4-5 Roppongi, Minato-ku, Tokyo 106-0032, Japan
TEL: +81-3-6910-0692
Principal Japanese Sources
Agency for Cultural Affairs, “AI and Copyright”
Agency for Cultural Affairs, “Checklist & Guidance on AI and Copyright”
Agency for Cultural Affairs, FAQs on Legal Issues in Cultural and Artistic Activities
Intellectual Property Strategy Headquarters, Intellectual Property Strategic Program 2026
Written and supervised by SAKURA Law Office, Managing Partner Kenshiro Michishita
This article provides general legal information based on the Japanese Copyright Act, materials published by the Agency for Cultural Affairs and the Cabinet Office, and generally accepted legal analysis as of September 24, 2026. The Agency for Cultural Affairs’ “General Understanding on AI and Copyright in Japan” is not legally binding, and Japanese case law remains limited on a number of copyright questions specific to generative AI. AI service terms, indemnity programs, training datasets and model specifications may also change over time. Specific matters should be assessed in light of the latest laws, judicial decisions and governmental materials, the contractual terms and technical specifications of the relevant AI service, the generation process, the intended use and the other facts of the particular case.