Dozens of lawsuits have been filed against AI companies over how models are trained and what they produce. Rulings have arrived in different courts and countries, and they do not all point the same way. For anyone who uses AI tools for work, it is hard to tell which headlines matter.
This article summarises the cases we could find in dated sources, as of 11 November 2026, and says plainly where we could not confirm something. It is a news summary. It is not legal advice, and the law differs by country. If a decision depends on copyright, speak to a qualified lawyer.
Key takeaways
- Courts have not given one answer. In the US, two June 2025 district court rulings found training use to be fair use on their facts, while a February 2025 ruling in Thomson Reuters v. Ross went the other way.
- The Anthropic authors’ class action was reported to have settled, with a reported figure of $1.5 billion, after the court distinguished training from keeping pirated copies.
- Outside the US, a UK court rejected most of Getty’s claims against Stability AI in November 2025, and a German court ruled for the music rights body GEMA against OpenAI the same month.
- In August 2026 a tracker reported rulings in the consolidated OpenAI litigation that dismissed some contributory infringement and trademark dilution claims.
- One source says no US federal appeals court has yet ruled on the core question of whether AI training is fair use. We could not confirm that independently.
Summary
Three questions run through the litigation. Is it lawful to train a model on copyrighted works without permission? Is it lawful to obtain those works from pirated sources? And can a model’s outputs infringe? The rulings we found answer the first two differently depending on the court and the evidence, and the third is less settled.
The best reading of the sources is that the law is developing case by case. A ruling that favours an AI company on one set of facts does not necessarily apply to another, and the judges themselves said so in at least one case. Appeals are still pending in some matters.
Key rulings at a glance
The table draws on Press Gazette’s tracker of AI legal rulings, last updated on 23 December 2025, and the other sources cited below. Dates are as stated by the sources.
| Case | Court | Outcome as reported | Date |
|---|---|---|---|
| Thomson Reuters v. Ross Intelligence | US court | Summary judgment that Ross’s use was not fair use; Ross has appealed | February 2025 |
| Bartz v. Anthropic | US district court | Training on lawfully acquired books held fair use; keeping a central library of pirated books not protected; class settlement followed | June 2025 |
| Kadrey v. Meta | US district court, Northern District of California | Partial summary judgment for Meta; judge said it does not set a wider precedent | June 2025 |
| Getty Images v. Stability AI | UK High Court | Getty lost its main claims; some trademark points won; permission to appeal granted on one issue | November 2025 |
| GEMA v. OpenAI | German court | GEMA won; OpenAI found to have breached copyright law and ordered to pay damages | November 2025 |
| Publishers v. Cohere | US District Court, Southern District of New York | Motion to dismiss denied | November 2025 |
| AlterNet and Raw Story v. OpenAI | US District Court | Complaint dismissed in its entirety | November 2024 |
US fair use rulings on training
Anthropic and Meta (June 2025)
Press Gazette reports that in Bartz v. Anthropic, Judge William Alsup found training use to be fair use, while keeping a central library of pirated books was found to infringe, opening the way to trial. In Kadrey v. Meta, Judge Vince Chhabria found Meta’s use to be fair use on that record but said the ruling does not set a wider precedent. A later summary written by an automated legal docket describes his reasoning as saying the plaintiffs made the wrong arguments and did not build a record of market harm. That page is machine-authored, so we treat its detail with caution, but it is consistent with the judge’s reported caveat.
Thomson Reuters v. Ross (February 2025)
The same tracker reports that Judge Stephanos Bibas granted summary judgment that Ross’s copying of Westlaw headnotes was not fair use, with some issues going to trial and Ross appealing. This case concerned a legal research tool, not a general chatbot, and that difference may matter.
Taken together, these decisions show why a one-line answer such as “AI training is fair use” or “AI training is infringement” is not supported by the sources. The outcome has depended on the use, the way the data was obtained and the evidence of market harm.
The Anthropic class settlement
Press Gazette says a settlement was reached in the Anthropic authors’ case in September 2025. The machine-authored docket page dated 1 September 2026 adds that the settlement was reported at $1.5 billion, roughly $3,000 per affected book, with preliminary approval in October 2025, final approval reported by mid-2026 and roughly 350 authors opting out. An AI Business article of 9 February 2026 also refers to a $1.5 billion settlement and says it set a litigation benchmark. A litigation tracker dated 29 August 2026 marks the Bartz class as settled. The approval details come from one source of uneven reliability, so treat them as reported.
News publishers and the OpenAI litigation
The consolidated litigation against OpenAI and Microsoft, brought by news organisations and others, continued through 2026. A post on the Chat GPT Is Eating the World tracker dated 7 August 2026 describes two rulings by Judge Stein in the consolidated case. It says they followed the plaintiffs’ voluntary dismissal of a “material contribution” theory of contributory infringement after the US Supreme Court’s decision in Cox Communications v. Sony Music Entertainment, as well as their trademark dilution claims. According to the post:
- The New York Times and Daily News did not show good cause to add new contributory theories against Microsoft.
- Their “material contribution” contributory claims against OpenAI and Microsoft, and their trademark dilution claims, were dismissed with prejudice.
- Ziff Davis’s corresponding active claims were also dismissed with prejudice, with claims about certain stayed models dismissed without prejudice.
- The news plaintiffs and Ziff Davis were to file amended complaints within fifteen days.
The machine-authored docket page of 1 September says direct infringement, vicarious infringement and DMCA claims were allowed to proceed. We could not confirm that from a second source, so read it with care. We did not find a final ruling on whether the underlying training was lawful in the news cases. The 29 August tracker lists the New York Times case without giving a status.
Cases outside the United States
Getty Images v. Stability AI (UK, November 2025)
Press Gazette reports that Getty lost its main claims, including secondary infringement, in the UK High Court. It won on some trademark points but not for additional damages. Getty was granted permission to appeal on the meaning of “infringing copy.”
GEMA v. OpenAI (Germany, November 2025)
The same tracker says a German court found for GEMA, holding that OpenAI breached copyright law and ordering it to pay damages. The page does not name the court.
These two outcomes differ in direction, which supports the point that jurisdiction matters. A decision in one country does not decide the question in another.
| Question | US rulings reported | Non-US rulings reported |
|---|---|---|
| Is training on copyrighted works lawful? | Fair use found in two June 2025 district rulings; not fair use in Ross (February 2025) | UK: Getty lost main claims; Germany: GEMA won |
| Do pirated source copies matter? | Yes, in the Anthropic ruling | Not addressed in the sources we opened |
| Can outputs infringe? | Cohere motion to dismiss denied; OpenAI cases continue | GEMA case involved findings against OpenAI |
Licensing deals
AI Business reported on 9 February 2026 that the New York Times and Amazon had a multiyear licensing deal valued at $20 to $25 million, and that Perplexity AI runs a revenue-sharing programme with news outlets. The article gives no dates for these deals and says settlement pressure was pushing OpenAI towards an agreement with the New York Times, but it does not say one was reached. We found no later confirmation. Licensing is a reminder that litigation is not the only path.
What is still undecided
- Appeals. Ross has appealed and plaintiffs in Kadrey were reported to be seeking review. One source says no federal appeals court has yet ruled on the core training question. We did not verify this independently.
- Output claims. Whether a model’s outputs infringe is still being litigated.
- Remedies and settlements. The Anthropic settlement is a reported benchmark, but other cases are at earlier stages.
- Scale. A 29 August 2026 litigation map lists roughly 125 entries and notes that some are duplicates or marked transferred, stayed or voluntarily dismissed, so counts vary.
What it means for people using AI tools
For most users, the main questions are practical. Can I use this output commercially? Could I be exposed if the output resembles someone else’s work? What do the vendor’s terms say?
- Read the vendor’s terms on ownership of outputs and any indemnity for IP claims. Terms differ by plan and vendor, and we have not reviewed them here.
- Do not paste in works you do not have rights to and expect a safe result.
- Check outputs for close matches to known works before publishing, especially for images, music and long text.
- Be transparent about AI use where your audience or clients would expect it.
Our guide on AI copyright basics for creators covers the everyday issues, and how to disclose AI use to your audience and how to cite AI-assisted work honestly cover transparency.
Practical steps
| Step | What to do |
|---|---|
| Know your use case | Internal drafts, public marketing and resale of outputs carry different levels of risk |
| Read the terms | Check output ownership, restrictions and any IP indemnity on your plan |
| Keep records | Note which tool and prompt produced significant content |
| Review before publishing | Check for close matches and unusual similarity to known works |
| Set a team rule | Put it in your policy; see how to write an AI use policy for your team |
| Get advice for high stakes | Ask a qualified lawyer where significant money or reputation is involved |
What to watch next
- Any appellate rulings in Thomson Reuters v. Ross or the Kadrey matter.
- The outcome of the UK appeal in Getty v. Stability AI.
- Whether the amended complaints in the OpenAI litigation, due after the August rulings, change the shape of the case.
- Further licensing deals between publishers and AI companies.
These are open questions, not predictions. This is general information, not legal advice.
Frequently asked questions
Is AI training legal?
The sources do not give a single answer. US courts have reached different outcomes on different facts, and courts in the UK and Germany reached different results. It depends on the country, the facts and the way the data was obtained.
Did Anthropic really settle for $1.5 billion?
Two sources we opened refer to a $1.5 billion settlement and a docket page gives further detail, but we could not open a court document. Treat the figure as reported.
Can I be sued for using AI output?
The sources we read are about claims against AI companies, not users. Risk depends on how similar the output is to protected work and how you use it. Ask a lawyer if the stakes are high.
Does it matter which AI tool I use?
Vendor terms differ on output ownership and indemnity, so yes. Read them for the plan you use.
Where can I follow the cases?
Litigation trackers such as the ones in our sources list cases and statuses, but they can lag or contain errors. Court records are the authoritative source.
Conclusion
As of November 2026, AI copyright law is unsettled and varies by jurisdiction. The reported rulings favour AI companies in some US training cases, go against them in others and diverge abroad, while settlements and licensing deals sit alongside the litigation. For users, the sensible approach is practical: read the terms, review outputs, disclose AI use and take advice when the stakes are high. Start with AI copyright basics for creators and revisit as the cases develop.
Sources
- Press Gazette, AI legal rulings tracker, last updated 23 December 2025, accessed 11 November 2026
- Chat GPT Is Eating the World, Judge Stein rulings in the OpenAI litigation, 7 August 2026, accessed 11 November 2026
- Chat GPT Is Eating the World, Map of copyright litigation v. AI companies in the United States, 29 August 2026, accessed 11 November 2026
- You.com, AI copyright case tracker (machine-authored), 1 September 2026, accessed 11 November 2026
- AI Business via Pulse, AI lawsuits in 2026: settlements, licensing deals, litigation, 9 February 2026, accessed 11 November 2026
