In Japan, content that a generative AI produces on its own is, as a rule, not protected by copyright. Only the parts where a person has made a "creative contribution" — through specific instructions or their own edits — may qualify as a copyrighted work (Copyright Act Art. 2(1)(i)).
At the same time, if an AI output resembles an existing work and was derived from it, it can infringe copyright regardless of whether AI was used. Copying works to train AI is generally permitted (Art. 30-4), but not where it would unreasonably prejudice the copyright owner's interests.
When This Becomes a Real Issue
- Your marketing team creates banner ads or packaging images with an image generator and wants to publish them as is.
- You want to deliver or license AI-written articles or code to a client as "your company's work."
- An employee has been typing another company's character or title names into prompts to create images.
- A rights holder demands that you take down images generated by users of your AI service.
Is AI Output Protected by Copyright?
The rule
A "work" is "a creatively produced expression of thoughts or sentiments" (Art. 2(1)(i)). An "author" is "a person who creates a work" (Art. 2(1)(ii)); AI has no legal personality, so it cannot be an author. AI output becomes a work only where the person using AI as a tool made a creative contribution.
"General Understanding on AI and Copyright in Japan," compiled by the Legal Subcommittee of the Copyright Subdivision of the Council for Cultural Affairs on March 15, 2024 (the "General Understanding"), identifies three factors:
- Volume and content of instructions: Detailed instructions that specifically set out creative expression make a finding of creative contribution more likely. Long instructions that only convey ideas do not.
- Number of attempts: The number of generations alone does not matter. Repeatedly revising instructions after checking outputs may, however, lead to protection.
- Selection among outputs: Merely choosing among outputs is not a creative contribution.
Example
An image generated from a simple prompt like "a bright beach scene for our summer campaign" is unlikely to be protected. If a designer then redraws the composition and adds colors and elements, the added parts are ordinarily protected. The unedited parts are assessed separately.
Practical points
If an output is not a work, copyright cannot stop competitors from copying it. A work created by an employee in the course of duty and published under the company's name is authored by the company unless a contract or work rules provide otherwise (Art. 15(1)). That rule assumes there is a work in the first place, so map out which parts are human-created before giving a client any warranty of rights.
Free Tool Related to This Article
Statute of Limitations Checker
Try our free simulator related to this topic.
Try for free →Infringement at the Generation and Use Stage: Similarity and Reliance
The rule
Japan's Supreme Court has held that a person who creates a similar work without knowing of the existing work is not liable for infringement (Sup. Ct., Sept. 7, 1978). In assessing adaptation, it asks whether the essential features of the original's expression are retained and can be directly perceived (Sup. Ct., June 28, 2001). The General Understanding applies these two requirements — similarity and reliance — to AI outputs as well.
What is distinctive about AI is reliance. The General Understanding sorts it into three cases:
| Situation | Reliance |
|---|---|
| The user knew the original | If the user creates something similar by, for example, inputting the image itself or a title or character name, reliance is found. |
| The user did not know it, but the AI was trained on it | If a similar output is produced, reliance is ordinarily presumed. |
| Neither the user knew it nor the AI was trained on it | The similarity is a coincidence and there is no reliance. |
Example
If an employee prompts with a popular character's name to make promotional art and posts it on social media, the infringement risk is high. Generating it only for internal review may fall within use in the process of considering exploitation with the owner's authorization (Art. 30-3). Publication or distribution, however, must be assessed separately.
Practical points
Caution
The Development and Training Stage: Article 30-4
The rule
Use of a work that is not for the purpose of enjoying the thoughts or sentiments expressed in it is permitted to the extent necessary (Art. 30-4). Data analysis is listed as an example (Art. 30-4(ii)), and AI training is understood to fall within it. The exception does not apply where the use would unreasonably prejudice the copyright owner's interests in light of the nature, purpose, and circumstances of use (proviso).
Where the exception may not apply
The General Understanding gives these examples:
- A concurrent purpose of enjoyment: Additional training intended to make the model output creative expression from the training data, such as deliberate overfitting. This can include fine-tuning on a small set of one creator's works for that purpose.
- The proviso: Copying data that is sold or licensed as a database organized for analysis, without using that offering. Technical measures such as robots.txt blocking AI crawlers, combined with signs that such a database will be sold, can point the same way.
- Pirated content: Knowingly collecting training data from sites hosting pirated copies should be strictly avoided. It increases the likelihood that the developer will be held responsible for later infringement.
Practical points
Style is treated as an idea, so a similar style alone is not infringement. A rights holder's statement objecting to AI training does not, by itself, trigger the proviso.
AI service providers should document their collection policies and data sources and adopt measures to prevent similar outputs. Providers whose models frequently generate infringing material are more likely to be treated as infringers themselves.
How to Use the Agency for Cultural Affairs' Two Documents
The General Understanding is not legally binding, but it is the current reference point for interpretation. The Agency's Copyright Division turned it into role-specific steps for developers, providers, users, and rights holders in its "Checklist & Guidance on AI and Copyright" (July 31, 2024). Basing an internal policy on the user section helps avoid gaps.
Beyond Copyright
Check AI-generated logos and product names against other companies' registered trademarks, regardless of whether they are copyrighted works. If outputs are used in product designs, check them against registered designs as well.
Images or voices resembling real people raise portrait and publicity rights issues. The Supreme Court has held that unauthorized use of a person's name or likeness aimed solely at exploiting its customer appeal infringes the right of publicity (Sup. Ct., Feb. 2, 2012).
What to Check in AI Service Terms
Important
Key points to check:
- Who owns outputs, and whether commercial use is allowed
- Whether your inputs are used for training, and how to opt out
- Restrictions on inputting others' works or titles
- The scope of indemnities and disclaimers if a third party claims infringement
If you provide an AI service, terms that discourage infringing use by customers are said to lower the risk of the provider being held responsible.
Building an Internal AI Policy
A workable policy rests on three pillars: approved tools, prohibited inputs, and review before publication. The Checklist & Guidance highlights checking service terms, training staff on copyright, and planning in advance how to respond to infringement. Training should make clear that whether an output is protected and whether it infringes someone else are separate questions.
Key point
When Another Company Alleges Infringement
First decide whether to suspend publication or distribution of the output in question. Then confirm internally which service and prompts were used and whether the creator knew the original. Similarity should be assessed on shared concrete expression, not on style or ideas.
If you can show the AI was not trained on the original, that weighs against reliance. You may also ask the AI provider for information on its training data. Establish the facts before responding, and decide carefully what, if anything, to concede.
Abroad: The U.S. Human Authorship Requirement
On January 29, 2025, the U.S. Copyright Office published its report on the copyrightability of AI outputs. It concluded that outputs can be protected where AI is used as a tool and a human determines the expressive elements, but that prompts alone are, at this stage, unlikely to be enough.
On March 18, 2025, the U.S. Court of Appeals for the D.C. Circuit affirmed the refusal to register a work listing an AI as author, holding that the Copyright Act requires human authorship. When registering in the United States, applicants must disclose AI-generated content that is more than de minimis and exclude it from the claim.
What You Can Do Now (Checklist)
- 1
Check service terms
List the generative AI services used in your company and check their terms on ownership, training use, and input restrictions.
- 2
Set input rules
Adopt a rule that, in principle, bans inputting other companies' titles, character names, or existing images.
- 3
Check for similarity
Add a step to search for similar existing works before publishing or delivering AI outputs.
- 4
Keep records
Save prompts and revision history, and record who edited which parts.
- 5
Clarify contracts
In client contracts, state whether AI is used and the scope of rights and warranties in outputs.
- 6
Set a response procedure
Designate a contact point and a procedure for suspending publication if infringement is alleged.
- 7
Document data policies
If you provide an AI service, document your data collection policy and measures to prevent similar outputs (Art. 30-4).
When to Talk to a Lawyer
- You receive a takedown demand or warning letter from a rights holder
- A client asks you to warrant rights in AI outputs you deliver or license
- You are deciding how to collect training data or draft the terms for your own AI service
- You are unsure whether fine-tuning on a particular creator's works falls outside Article 30-4
- You plan to use outputs in the United States or other countries and need to reconcile different rules
You can reach us through our inquiry page.
FREE CONSULTATION
Internet Issues: your first 30 minutes are free
- ・Handled by Yusuke Abe, admitted in Japan and New York
- ・Online (Zoom etc.) or at our office in Otemachi, Tokyo
- ・In English or Japanese
Frequently Asked Questions
Are images or text created with AI protected by copyright in Japan?
As a rule, content generated autonomously by AI is not protected. Parts where a person gave instructions that specifically set out creative expression, or made creative edits to the output, can be protected. Simply generating many times or choosing among outputs is not considered enough.
Is it infringement if an AI output resembles an existing work?
It can be, if both similarity and reliance are found. Even if the user did not know the original, reliance is ordinarily presumed if the AI was trained on it. A similar style alone is not infringement.
Is it illegal to use others' works to train AI in Japan?
As data analysis, it is generally permitted without permission (Copyright Act Art. 30-4). The exception does not cover additional training meant to output the training data's expression, or uses that unreasonably prejudice the copyright owner's interests. Knowingly collecting pirated content should be strictly avoided.
If an AI service's terms say "you own the output," is it our copyrighted work?
The terms govern rights between the provider and the user. They cannot create copyright in output that is not a work. Nor do they guarantee that the output does not infringe others' rights, so that must be checked separately.
Does the company own what employees create with generative AI?
For parts that qualify as works, the company is the author if the requirements are met — for example, the work was made in the course of duty and published under the company's name — unless otherwise agreed (Copyright Act Art. 15(1)). Parts that are not works have no copyright at all. An internal policy on handling outputs and keeping records is the safer approach.