TL;DR (Updated May 2026)
- AI output alone is NOT copyrightable in Japan. Copyright requires "creative expression of thoughts or feelings" by a human (Copyright Act Art. 2(1)(i)).
- Detailed human prompting + selection + editing can produce copyrightable work — but only the human-contributed portion is protected.
- AI training on copyrighted works is lawful under Art. 30-4 (Japan's information-analysis exception), but this does NOT shield you from infringement at the output stage.
- Generating outputs similar to existing works is infringement (Arts. 21-28) if "dependency" and "substantial similarity" are met.
- Corporate use requires written guidelines: confidential-data ban, similarity checks, contractual IP allocation with clients/vendors.
Quick Reference: Is My AI Output Copyrightable?
| Scenario | Copyrightable in Japan? | Why |
|---|---|---|
| One-shot prompt → use raw output | ❌ No | Lacks human creative contribution (ACA 2023) |
| Long, specific prompt → iterate 5+ times → select & edit | ✅ Likely yes (human portion) | Selection, arrangement, refinement = creative work |
| AI generates near-replica of existing work | ❌ No, and infringes | Dependency + substantial similarity → Arts. 21-28 |
| AI-assisted translation reviewed by human | ✅ Yes for human edits | Editorial contribution protectable |
| Photo edited with Generative Fill (Photoshop AI) | ⚠️ Depends on extent | Original photo: yes; AI-filled portion: case-by-case |
---
Free Tool Related to This Article
Statute of Limitations Checker
Try our free simulator related to this topic.
Try for free →Statutory Framework
Under Japan's Copyright Act, works require "creative expression of thoughts or feelings" (Art. 2(1)(i)). AI-generated content may not qualify as a copyrightable work due to lack of human creative contribution (Agency for Cultural Affairs 2023 guidance). However, detailed human direction may create copyrightable output. AI training on copyrighted works is generally lawful for information analysis (Art. 30-4), unless it unduly harms the copyright holder's interests. Risk: AI-generated content may infringe existing copyrights through similarity (Arts. 21-28). Key issues: proving "dependence" on training data. Corporate guidelines should address usage scope, confidential data restrictions, similarity checking, and contractual copyright allocation.
International Regulatory Developments
EU AI Act (Enacted 2024)
The EU AI Act, enacted in 2024, imposes transparency obligations on providers of general-purpose AI systems. Under Article 53, providers must disclose a summary of copyrighted content used in training data. AI-generated content must also be labeled as AI-generated. These rules affect Japanese companies operating in European markets.
U.S. Copyright Office AI Guidance
The U.S. Copyright Office issued guidance in 2023–2024 reaffirming that human creative authorship is an indispensable requirement for copyright protection. Purely AI-generated content is not protectable, but human selection, arrangement, or modification of AI output may qualify. The 2024 Copyright Office report explicitly states: "AI cannot be an author under U.S. copyright law."
China: Beijing Internet Court Recognizes AI-Assisted Copyright (2023)
In 2023, the Beijing Internet Court issued a landmark ruling recognizing copyright in an AI-generated image, where the user had made creative decisions through prompt iteration, compositional choices, and repeated refinements. This decision — while not binding on Japanese courts — is widely cited as evidence that prompt-level creative involvement can support a copyright claim.
Dependency and Similarity: Decision Framework
Whether AI-generated content infringes an existing work hinges on two key requirements: "dependency" (依拠性) and "substantial similarity" (類似性).
| Requirement | Standard | Example |
|---|---|---|
| Dependency (依拠性) | Was the existing work included in training data? Did the AI "base" its output on it? | Instructing AI to mimic a specific author's style — likely dependent if that author's works were in training data |
| Substantial Similarity (類似性) | Can the essential characteristics of the original expression be directly perceived? Idea vs. expression: only expression is protected | Shared sentence structure, word choices, or distinctive phrasing overlapping with the original |
| Proving Dependency is Difficult | Training data contents are typically undisclosed; proving dependency is practically challenging for users | Users rarely have access to AI training data manifests |
| Pre-publication Verification | Companies and users are expected to conduct similarity checks before publishing AI output | Use plagiarism checkers, reverse image search, or similarity detection tools |
How to Establish Creative Contribution Through Prompting
To support a copyright claim over AI-generated content, document your creative involvement from the prompting stage.
1. Provide Detailed, Specific Instructions
Rather than generic prompts like "write an article about X," specify structure, tone, key arguments, and exclusions. The more specific and deliberate the instruction, the stronger the case for human creative judgment.
2. Document an Iterative Revision Process
Don't use first-draft output directly. Run multiple prompt iterations, compare alternatives, and log your choices. The Agency for Cultural Affairs specifically cited "repeated modifications" as supporting creative contribution — keeping a log of this process serves as evidence.
3. Add Human Selection, Arrangement, and Editing
Selecting among multiple AI outputs, reordering sections, and rewriting in your own voice all constitute human editorial work that strengthens the case for copyrightability.
4. Save Prompts and Outputs Together
Preserve prompts paired with their corresponding outputs to demonstrate what instructions produced what results. This record becomes critical evidence if copyright ownership is later disputed.
Case Studies (Practical Patterns)
Case 1: SaaS Company Using ChatGPT for Marketing Copy
A Tokyo-based SaaS company uses ChatGPT to draft blog posts. The marketing lead provides a 300-word brief specifying tone, structure, target keywords, and key arguments, then revises the AI draft three times before publishing. Result: The human-contributed portions (structure, key arguments, revisions) are copyrightable. The raw AI-generated sentences in between may not be. Best practice: treat the published post as a "compilation work" and document the human input.
Case 2: Game Studio Using Midjourney for Concept Art
A game studio uses Midjourney to generate 200 concept images. An art director selects 15, edits them in Photoshop with significant human modifications (composition, color grading, additional elements), then uses them as references for human-drawn final art. Result: The original Midjourney outputs are not copyrightable on their own, but the edited versions and the final human-drawn art are. Risk: if any AI output resembles a known artist's style closely, similarity claims may arise — verify with reverse image search before commercial use.
Case 3: Law Firm Drafting Contracts with AI
A law firm uses an AI legal research tool to generate first drafts of standard contracts, which attorneys then review, customize, and finalize. Result: The customized final contract reflects the attorney's professional judgment and creative arrangement — copyrightable as a work product. The AI's initial draft, isolated, would not be. Confidentiality note: client data input to public AI services may violate attorney-client privilege; on-premise or contract-protected AI is required.
Frequently Asked Questions
Q1. If I use ChatGPT to write an article, can I claim copyright?
Only over the parts you creatively contributed. Pure AI output without significant human direction is not copyrightable in Japan. To strengthen your claim, document your prompt design, iteration log, and editorial choices.
Q2. Can I be sued if ChatGPT writes something similar to a copyrighted novel?
Yes. The user (or company deploying the AI) bears responsibility for infringing output. Substantial similarity + dependency on the original work = infringement under Arts. 21-28, regardless of whether you "knew" the AI was trained on that work.
Q3. Is OpenAI / Anthropic / Google liable instead of me?
Provider liability is unsettled. The Agency for Cultural Affairs has discussed shared responsibility, but currently the user/deployer carries primary risk in Japan. Provider terms of service typically disclaim liability and require users to indemnify.
Q4. Does Article 30-4 protect me from any AI copyright lawsuit?
No. Art. 30-4 only covers the training stage (ingesting copyrighted works to train the model). It does NOT cover output that infringes specific works. Two separate analyses: (1) training = generally lawful; (2) output = traditional infringement test applies.
Q5. What if AI generates an image identical to a famous photograph?
That's textbook substantial similarity. The user is liable for infringement (reproduction right Art. 21, public transmission Art. 23). Damages may include lost licensing fees and reputational harm. Mitigation: similarity-check generated images against major stock libraries before commercial use.
Q6. Can I copyright a prompt itself?
Possibly, if the prompt is sufficiently creative as a literary expression. Short utilitarian prompts ("write a blog post about cats") are likely not copyrightable. Long, structured prompts with distinctive phrasing may qualify as literary works.
Q7. What should our company's AI usage policy include?
At minimum: (1) prohibited input categories (personal data, trade secrets, client confidential information); (2) approved AI tools list; (3) similarity-check requirement before commercial publication; (4) attribution and disclosure rules; (5) contractual flow-down to vendors and clients; (6) recordkeeping of prompts and outputs.
Q8. Are there pending court cases in Japan that may change this?
Yes — several infringement claims against AI training and outputs are working through Japanese courts in 2025-2026. Watch the Tokyo District Court IP division. We update this article quarterly; next update scheduled August 2026.
Practical Compliance Checklist for Japanese Companies
- ☐ Document who in your organization may use AI (named roles, not "anyone")
- ☐ Approve specific AI tools after legal review of their terms of service
- ☐ Prohibit input of personal data (Personal Information Protection Act) and trade secrets
- ☐ Mandate human review + editorial intervention before publishing AI output
- ☐ Run similarity checks against major content databases before commercial use
- ☐ Update client contracts to allocate AI-generated IP rights explicitly
- ☐ Maintain prompt + output logs for 5+ years (statute of limitations buffer)
- ☐ Train staff on Art. 30-4 vs. output infringement (two-stage analysis)
Important Notes
- This article is for general information, not legal advice for a specific case. Consult a Japanese qualified attorney for matters involving actual disputes, contracts, or compliance decisions.
- Last updated: May 11, 2026. Japan AI copyright law is rapidly evolving — ACA guidance, METI guidelines, and pending court decisions may shift positions.
- Practice area: Internet law, intellectual property, corporate compliance.
- For consultation regarding AI-related copyright issues, defamation, or platform liability in Japan, contact our office through the inquiry form.