Courtroom to Contract: AI Lawsuits in 2026, Settlements and Licensing Deals
Slow court cases, big settlements and new licensing deals: where AI lawsuits stand in 2026 and what they mean for small businesses, with a practical risk check.
A small business owner using a chatbot to draft blog posts, or an image tool to make ads, keeps seeing headlines about billion-dollar settlements and landmark rulings. It is fair to wonder whether any of it reaches a business that simply pays a monthly subscription. Mostly, AI lawsuits are fights between large AI vendors and large rights holders. Their outcomes still flow downstream, into what tools cost, what their terms allow and what is safe to publish. In 2026 the pattern became clearer: slow court cases, faster settlements, and a licensing market growing in the gaps.
Where the major AI lawsuits stand
The case that framed the debate is moving slowly. The New York Times sued OpenAI and Microsoft in December 2023, arguing that its articles were used without permission or payment to train ChatGPT. In March 2025, Judge Sidney Stein rejected OpenAI’s request to dismiss the case and let the main copyright claims proceed, as NPR reported via Michigan Public. OpenAI’s defense rests on fair use, the US doctrine that allows some unlicensed use of copyrighted material.
Other courts have reached decisions, and they do not all point the same way. In June 2025, two federal judges in California ruled within days of each other. According to an analysis by Authors Alliance, Judge William Alsup found that training Anthropic’s models on books was fair use but that building a central library of pirated books was not, while Judge Vince Chhabria ruled for Meta on the record before him and floated a new “market dilution” theory that future plaintiffs might use.
The Anthropic case then ended in money. A federal judge gave final approval in July 2026 to a $1.5 billion settlement covering almost 500,000 books, about $3,000 per work, Thurrott reported.
In September 2026 came what law firm Haynes Boone describes as apparently the first federal appeals ruling on fair use in AI training. The Third Circuit held that Ross Intelligence’s copying of Westlaw headnotes to train a legal research tool was not fair use. The tool was not generative, which limits the reach, but the court recognized licensing content as AI training data as a real market.
Outside the US, the picture is mixed. The UK High Court ruled in November 2025 that Stability AI’s model was not an infringing copy of Getty Images’ photos, a decision Getty can appeal, according to AIPPI. In Germany, the Munich Regional Court found that OpenAI infringed copyright when its models memorized and reproduced song lyrics managed by the collecting society GEMA, a ruling that is not yet final, as summarized by the law firm Grünecker.
Why settlements and licensing deals keep multiplying
Courts are slow and uncertain, so both sides have reasons to negotiate. Warner Music Group settled with the AI music service Suno in November 2025 and turned the dispute into a licensing partnership, with Suno planning licensed models for 2026, Music Business Worldwide reported. A month later Disney announced a three-year agreement letting OpenAI’s Sora video tool generate its characters, alongside a $1 billion investment in OpenAI, according to NBC.
The reasonable reading is that training data is becoming something vendors pay for, through settlements or contracts. Over time, those costs are likely to appear somewhere, possibly in subscription prices or in tighter terms about what users may generate.
A risk check for small businesses
The table below maps the cases above to everyday situations. It is general information drawn from those sources, not legal advice.
| What the business does | What the cases suggest | Practical step |
|---|---|---|
| Uses a mainstream chatbot for drafts | Lawsuits so far target vendors; the Munich court placed liability for infringing outputs on the provider | Keep using it, and edit drafts into original work |
| Publishes output that closely reproduces lyrics, articles or famous characters | Memorized and reproduced works were found infringing in Munich | Check outputs for recognizable text or characters before publishing |
| Builds a custom tool trained on others’ content | Copying to build a competing product was not fair use in the Ross case; pirated sources were not protected in the Anthropic case | Train only on content the business owns or has licensed |
| Has valuable content of its own | Courts and deals now treat training data as a licensable market | Decide deliberately which AI crawlers may read the website |
| Chooses a new AI vendor | Licensing and settlements are becoming normal costs for vendors | Prefer vendors that explain their training data and commercial terms clearly |
One step for this month
Open the terms of service of the AI tool the business uses most and find two things: who owns the output, and whether the vendor offers any protection if that output is challenged. Ten minutes of reading answers more than a year of headlines.
Related guides cover AI crawler rules for business websites, who owns AI-generated content and the legal questions agencies now face. More analysis sits under AI for Business Owners.
If training data keeps getting more expensive, who will end up paying for it: the AI companies, or the small businesses that rely on their tools?
In this article
Light Rules, Faster Work: Getting AI Governance Right Without Slowing Down
Bans push AI use out of sight and free-for-alls lose track of it. A lighter path to AI governance for small businesses, with a five-day starter pla…
