The Wikimedia Foundation asked a US court not to treat AI and copyright as all-or-nothing. Here is what it said, where AI copyright law stands, and what creators should do now.
On October 30, 2026, the Wikimedia Foundation, which runs Wikipedia, filed an amicus brief in the OpenAI copyright litigation in New York. It backs neither side. It says AI training may qualify as fair use, but chatbots that reproduce works for users deserve firmer limits. It warns that a ruling too narrow or too broad could harm free knowledge and the open web.
AI copyright is now a fight over the open web itself. On October 30, 2026, the Wikimedia Foundation, the nonprofit that runs Wikipedia, urged a federal court to avoid an all-or-nothing ruling in the biggest US generative AI copyright case. The foundation warns that a bad AI copyright ruling, in either direction, could weaken free knowledge online.
Wikimedia filed an amicus brief, or friend-of-the-court brief, in In re OpenAI, Inc. Copyright Infringement Litigation. It is a consolidated case in the Southern District of New York (No. 1:25-md-03143). Publishers such as The New York Times are suing OpenAI and Microsoft there. Wikimedia sided with neither. Instead, it asked the court to treat AI training and AI outputs as separate questions.
This guide explains Wikimedia’s argument. It then maps the wider landscape of AI and copyright: who owns AI-generated content, what courts say about AI training data and fair use, and what creators and businesses should do next. It also covers what Nigerian and African creators need to know.
What Wikimedia Told the Court About AI Copyright
A Middle Path on AI Training and Outputs
According to coverage of the filing, Wikimedia is urging courts to separate two things. The first is training large language models on copyrighted works. The second is AI systems reproducing those works for users. In its view, training may qualify as fair use. Chatbots that regurgitate protected content deserve a firmer line.
Wikimedia says both sides push the court toward overbroad answers. The news plaintiffs argue that training and retrieval on their articles are not fair use. OpenAI argues that its pretraining was transformative and that verbatim outputs are rare. Its expert reported 24 verbatim outputs in 20 million chats. Both sets of briefs were filed on October 23, 2026.
Two Risks, One Warning
Wikimedia’s warning cuts both ways. A narrow ruling and a broad ruling each threaten the open web, but through different routes.
| IF THE COURT RULES… | WHAT COULD HAPPEN | WHY IT HURTS OPEN KNOWLEDGE |
|---|---|---|
| Too narrowly (for publishers) | Training itself is treated as infringement. Rights holders can block transformative uses. | Wikipedians could lose useful AI-based tools. Public content that open-source AI needs becomes harder to use. |
| Too broadly (for AI companies) | Blanket fair use, even for verbatim outputs. AI firms could skip licensing terms. | Publishers lock more content behind barriers. Attribution licenses lose force. Fewer people may share knowledge online. |
Why an AI Copyright Ruling Could Threaten the Open Web
Open Licenses Only Work if Everyone Follows Them
Wikipedia is written by volunteers and shared under open licenses that require attribution. In its Diff blog post, the foundation says AI companies that use Wikipedia content through retrieval-augmented generation (RAG) or similar techniques must honor those requirements. A blanket fair use ruling would let developers ignore them. That would break the link between human authors and the audiences who read their work.
The foundation also offers a licensed route. Wikimedia Enterprise, launched in 2021, sells high-volume access to Wikipedia content for large technology firms. In short, a fair path to use open knowledge already exists.
The Walls Around the Web Are Already Going Up
Other platforms are not waiting for the courts. Reddit says it is shifting to a logged-in experience to protect its content from AI scraping; read our report on Reddit’s logged-in experience and falling search traffic. Reference publishers are suing too. Britannica and wikiHow have both sued OpenAI over training, retrieval, and alleged verbatim outputs.
The strain is also technical. Wikipedia’s own entry on the foundation reports that Wikidata suffered a major outage on October 6, 2026. The foundation blamed OpenAI agents that sent thousands of queries to the service, and OpenAI said it would help investigate. Treat that account with care, since it comes from a public wiki entry rather than a court record.
Why Wikimedia’s Voice Matter’s
Most outside briefs in this case come from for-profit companies and trade groups. The Association of American Publishers, for example, has filed briefs opposing broad fair use protection for AI developers. Wikimedia’s General Counsel, Stephen LaPorte, argues that the fair use debate must include the whole online information ecosystem, not only commercial players.
“defend the free and open human knowledge that Wikipedia depends on”— Bernadette Meehan, Wikimedia CEO, on why the foundation took a position
Where AI Copyright Law Stands in 2026
There is no single answer yet. AI copyright case law is split, and no US appeals court has ruled on generative AI training. The table below shows the cases that shape AI copyright policy today.
| CASE | COURT AND DATE | WHAT HAPPENED | WHY IT MATTERS |
|---|---|---|---|
| Bartz v. Anthropic | N.D. Cal.; June 2025; settlement approved July 20, 2026 | Training on lawfully acquired books was fair use. Pirated library copies were not. Settled for about $1.5 billion. | How data is acquired matters. |
| Kadrey v. Meta | N.D. Cal.; June 2025 | Fair use for Meta on that record, because authors showed no market harm. Claims over torrent seeding remain. | Market-harm evidence can decide the outcome. |
| Thomson Reuters v. Ross | Third Circuit; Sept 29, 2026 | Training a competing legal search tool on Westlaw headnotes was not fair use. | First US appeals ruling on AI training, but the tool was not generative. |
| In re OpenAI Copyright Litigation | S.D.N.Y. MDL 1:25-md-03143; Oct 2026 | Summary judgment briefing on training, retrieval, and outputs. | Wikimedia’s brief targets this case. |
| Disney, Universal, and Warner Bros. v. Midjourney | C.D. Cal.; filed June 2025 | Studios allege image outputs copy famous characters. Discovery fights continued in July 2026. | Tests output infringement for AI images. |
| Getty Images v. Stability AI | UK High Court, late 2025; US case in N.D. Cal. | In the UK, Getty lost its secondary copyright claim but narrowly won on trademark over watermarks. Permission to appeal was granted. | Shows the evidence hurdles in training claims. |
| GEMA v. OpenAI | Munich Regional Court; Nov 2025 | First European court decision to find infringement in training an AI model. | Europe may be less friendly to unlicensed training. |
| ANI v. OpenAI | Delhi High Court (interim) | Training-time storage may be fair dealing. Outputs must be judged separately. | Another court that splits training from output. |
AI Sphere Hub analysis. Courts in the US and India are already separating training from output. Wikimedia’s two-stage test fits that trend. The strongest near-term legal risks sit with how data was acquired and with outputs that copy protected content.
AI Training Data Copyright and Fair Use
Fair Use and Transformative Use in AI Training
Fair use is a US legal doctrine. Judges weigh the purpose of the use, the nature of the work, how much was taken, and the effect on the market. In AI model training copyright cases, two questions dominate. First, is the use transformative? Second, does the AI product substitute for the original work and hurt its market?
AI developers say training is transformative because models learn patterns rather than copy text. Publishers answer that unlicensed AI training builds substitutes for their work and takes markets that belong to them. The Third Circuit’s Ross decision backed the second view for a competing legal tool. It left the generative AI fair use question open.
Scraping, Piracy, and AI Dataset Copyright
Where training data comes from has become a major liability issue. Pirated shadow-library books drove Anthropic’s roughly $1.5 billion settlement, which covered about 500,000 books at around $3,000 per work. Meta won on fair use, yet claims over torrent seeding survive. Meanwhile, Getty’s case against Stability AI rests on the LAION-5B dataset.
Web scraping and copyright disputes are growing too. Reddit has sued several scraping companies, and Britannica has sued over retrieval-based answers. Scraping copyrighted content is not automatically illegal, but it raises copyright, contract, and terms-of-service risks.
Training Data Licensing and AI Copyright Exceptions
Training data licensing is growing fast. Reddit licenses its data to OpenAI and Google. In music, Warner settled with Suno and Udio, and Universal settled with Udio, as industry reporting describes. A startup called Klay signed deals with all three major labels for a model trained only on licensed catalog music.
| REGION | TRAINING EXCEPTION OR RULE | STATUS IN 2026 |
|---|---|---|
| United States | Fair use, decided case by case | Split district rulings; no appellate ruling on generative training |
| European Union | Text and data mining exception with rights-holder opt-out; AI Act training-data summaries | AI Act transparency duties phasing in; Munich court found infringement in GEMA case |
| United Kingdom | Narrow non-commercial research exception | March 2026 report dropped the preferred opt-out exception; no immediate reform |
| India | Fair dealing | Interim Delhi ruling leaned toward training as fair dealing |
| Nigeria | No AI-specific provisions in the Copyright Act 2022 | No Nigerian court ruling yet on AI training |
AI Copyright Ownership: Who Owns AI-Generated Content?
The Human Authorship Rule for AI Authorship
US law requires human authorship for copyright. On March 2, 2026, the Supreme Court declined to hear Thaler v. Perlmutter. That left in place a 2025 appeals ruling that AI-generated art with no human creative input cannot be copyrighted. AI authorship alone does not count, and the same logic applies to patents: AI cannot be named as an inventor under US patent law.
The ruling did not say how much human input is enough. A pending case, Allen v. Perlmutter in Colorado, may help. An artist there challenges the refusal to register an image made with Midjourney using more than 600 prompts.
Prompt Copyright and the Copyright Office AI Guidance
The US Copyright Office published its AI guidance on copyrightability on January 29, 2025. It concluded that prompts alone do not give users enough control over the output to make them authors. Human-authored expression that remains visible in the output can be protected. So can creative selection and arrangement, and creative modifications of AI output. Originality still matters: the work must reflect real human creative choices.
AI-Assisted Work Copyright by Content Type
| CONTENT TYPE | WHAT MAY BE PROTECTED | WHAT USUALLY IS NOT |
|---|---|---|
| AI-generated text (AI writing copyright) | Your own passages, heavy edits, original structure | Raw, unedited AI output |
| AI-generated art and images (copyright protection for AI art) | Creative selection and arrangement; your modifications | The raw AI image itself |
| AI-generated music | Human lyrics, melodies, arrangement, performance | Fully generated tracks with no human authorship |
| AI-generated video | Human script, editing, sequencing, added footage | Raw generated clips |
| AI code and AI software copyright | Human-written code, original design, and arrangement | Code the model wrote with no human creative control; also check open-source license duties |
Real examples show the split. The Office registered the selection and arrangement of AI-generated images in a comic book, but not the images themselves. Since 2023, it has registered hundreds of works that include AI material, covering only the human contribution.
AI Copyright Registration and Disclosure
When you register AI-assisted work, disclose the AI-generated material and exclude it from your claim. Claim only your human-authored contribution. This covers registering AI-generated work in a mixed project: you describe what you did, and the Office decides what to protect. Keep drafts and process records ready in case the examiner asks.
AI Likeness Rights, Voice Cloning, and the Right of Publicity
Copyright does not protect a person’s voice or face as such. It protects recordings, compositions, and other fixed works. Voice cloning and AI likeness rights therefore rely on the AI right of publicity, state laws, and contracts.
The Copyright Office’s first AI report, from July 2024, covered digital replicas and copyright. Congress has reintroduced the NO FAKES Act, but it had not passed as of April 2026 reporting. Tennessee’s ELVIS Act has been in force since July 2024. Meanwhile, the Warner–Suno deal gives artists control over how their name, voice, image, and compositions are used in AI tracks. Artist rights and AI are increasingly set by contract rather than by statute.
AI Intellectual Property: Patent vs Copyright vs Trademark
| RIGHT | WHAT IT PROTECTS | AI ISSUE TODAY |
|---|---|---|
| Copyright | Original creative expression | Needs human authorship; training and output disputes |
| Patent | New, useful inventions | AI cannot be named as inventor in the US (Thaler v. Vidal) |
| Trademark | Brand names, logos, and marks | AI outputs showing distorted watermarks or logos can cause confusion claims (Getty) |
| Right of publicity | Name, image, voice, likeness | Voice cloning and deepfakes; state laws and contracts |
These areas together make up AI intellectual property and AI IP law. Artificial intelligence intellectual property questions rarely fit into one box, so check each right separately.
Commercial Use: AI Content Licensing and Tool Terms
Commercial use of AI-generated images and AI-generated content depends on three things. First, the AI-generated content terms of service. Second, whether the output copies protected material. Third, whether anyone can own the result. Each AI tool copyright policy differs and changes often, so read the current version for your plan.
| TOOL | USED FOR | WHAT TO CHECK BEFORE COMMERCIAL USE |
|---|---|---|
| ChatGPT copyright | Text, code, images | Output ownership clause, plan tier, data settings |
| Midjourney copyright | Images, video | Commercial-use rules by plan; studio lawsuit exposure |
| DALL-E copyright | Images | Output rights and content policy under current terms |
| Stable Diffusion copyright | Images (open model) | Which model version and license you use; dataset disputes |
| Sora copyright | Video | Terms on likeness, characters, and labeling |
An AI art copyright policy on a site or app is a contract, not a law. It can grant you rights the tool maker holds. It cannot create copyright the law does not recognize.
License AI-Generated Content and AI Image Licensing
Because raw AI output may lack copyright, exclusive licensing is hard. Licensing AI-generated content works best when you license the human-authored layer: your edits, composition, and original additions. For AI image licensing and AI stock content copyright, check each agency’s contributor rules, because many restrict or label AI submissions. For AI music licensing, prefer tools with clear licensed training data and written permissions. AI content ownership for business should also be spelled out in contracts, including who owns AI-assisted deliverables and who carries the risk if they infringe.
AI Content Attribution and Content Usage Rights
AI content attribution matters in two ways. When you use open material such as Wikipedia, you must follow its license, which means giving credit. When you publish AI content, disclosure builds trust. The EU AI Act’s Article 50 transparency rules, which cover labeling of synthetic content, were scheduled to become enforceable on August 2, 2026. AI content usage rights also depend on the platform where you post, so read its AI policy.
AI Copyright Rules for Creators and Businesses
| WHO YOU ARE | MAIN RISK | WHAT TO DO |
|---|---|---|
| Content creators (AI copyright law for content creators) | Copied passages; no exclusive rights in raw output | Edit heavily, add original work, keep drafts |
| Bloggers (AI copyright rules for bloggers; copyright rules for AI-generated blog posts) | Near-verbatim text; weak trust signals | Check for plagiarism, cite sources, quote sparingly |
| Freelance writers (AI copyright for freelance writers) | Contracts that require original work and AI disclosure | Read the contract; disclose AI use; deliver human-edited work |
| YouTubers (AI copyright for YouTubers) | AI voices, faces, and music; platform labels | Use licensed music; avoid cloned voices; follow platform disclosure rules |
| Musicians (AI copyright for musicians) | Likeness, sample, and lyric issues; weak protection for generated tracks | Document human parts; use licensed tools; register compositions |
| Graphic designers (AI copyright for graphic designers) | Raw AI art is not protectable; client expectations | Add original design layers; put AI use in the contract |
| Businesses and small businesses (AI copyright for businesses; AI copyright guide for small businesses) | Ownership gaps; infringement exposure | Set an AI policy; vet tools; require human review |
| Artists (AI copyright guide for artists) | Training on their work; imitation | Register key works; watch lawsuits and licensing options |
| Publishers (AI copyright guide for publishers) | AI submissions; scraping of archives | Add AI clauses to contracts; explore licensing deals |
AI-Generated Content and SEO Copyright Risks
Copyright and search visibility are linked. Near-verbatim copies can draw takedown notices, and removed pages lose traffic. Copied passages also weaken trust signals. At the same time, AI answers are cutting clicks, as our Reddit report shows, so original, well-cited pages matter more. Write facts in your own words, name your sources, and add real expertise.
Legal Risks of Publishing AI-Generated Content
The main legal risks are simple. Output may reproduce protected text or images. You may have no exclusive rights in raw AI content. You may breach a tool’s terms or a client contract. And you may use someone’s voice or likeness without permission. AI plagiarism, AI-generated content plagiarism, and copyright infringement by AI all trace back to one habit: publishing without review.
AI Copyright Checklist: How to Avoid Infringement and Protect AI-Assisted Work
- Read the tool’s terms and your planCheck output ownership, commercial-use rules, and what the tool says about copyright claims.
- Keep a human in the creative loopEdit, rewrite, arrange, and add original expression. Raw AI output alone may have no protection.
- Document your processSave prompts, drafts, versions, and edits to show your contribution.
- Check outputs for copyingRun plagiarism checks on text, reverse image searches on images, and review code for copied licensed material.
- Avoid protected names, characters, and likenessesDo not prompt for named characters, brands, living artists’ works, or a real person’s voice or face without permission.
- Disclose AI use where requiredTell clients, publishers, platforms, and the Copyright Office when AI material is involved. This is how to disclose AI-generated content properly.
- License what you canPrefer tools and datasets with clear licensing. Keep licenses and invoices. This is a core part of how to protect AI-generated content.
- Register the human contribution and review quarterlyRegister important human-authored work, and update your policy often. These are best practices for AI copyright compliance.
What the AI Copyright Fight Means for Nigerian and African Creators
Nigerian law requires a human author. Under the Copyright Act 2022, an author does not include a machine, so works made purely by AI are not protectable. The Act has no specific AI provisions, according to an April 2026 legal survey. Commentators note that AI-assisted work may still qualify where a person adds meaningful transformation and creative development.
No Nigerian court has ruled yet. The question of AI-generated works has not reached a Nigerian court. The Nigerian Copyright Commission enforces the Act. Until clearer guidance arrives, Nollywood producers, Afrobeats artists, designers, and writers should document human input and disclose AI use in contracts.
The open web matters more where data is costly. AI Sphere Hub analysis:Free reference sources like Wikipedia help students, developers, and small publishers who cannot afford many subscriptions. If courts push publishers to lock content away, those users lose first. Respecting attribution and licensing also protects contributors from Africa and elsewhere who write for open projects.
Treat foreign rulings as signals, not rules. US and UK decisions do not bind Nigerian courts. Still, they shape tool terms, platform policies, and client expectations. Follow the cases listed above, and ask a Nigerian IP lawyer before you rely on any of them.
What Happens Next in AI Copyright Reform
Several decisions will shape AI copyright legislation and reform over the next year. Here is what to watch:
- The OpenAI multidistrict case: Summary judgment rulings on training, retrieval, and outputs.
- More appeals: Whether another US appeals court decides generative AI training fair use.
- Allen v. Perlmutter: How much human input makes AI-assisted images registrable.
- UK reform: The government’s March 2026 report chose no immediate changes and signaled a review of the computer-generated works rule, section 9(3). It followed an AI copyright policy consultation in which only 3% of respondents backed the government’s preferred opt-out.
- Digital replica laws: Whether Congress passes the NO FAKES Act.
- EU enforcement: How transparency and labeling duties under the AI Act are applied.
Stay Ahead of AI Copyright and Policy News
AI Sphere Hub explains AI law, policy, and creator rights in plain language, with an Africa and Nigeria lens.