AI Lawsuit Tracker: 248 Cases Against 76 Companies and What the Data Shows

A public database now counts 248 lawsuits against 76 AI companies across 14 jurisdictions. The AI Lawsuit Tracker, described on its page as an editorial database of AI-related lawsuits, rulings and settlements "built from public court records," was last updated on October 8, 2026 and surfaced on Hacker News on October 11. For anyone building on AI models, the useful question is what the pile of cases tells you about legal risk, not just how large it is.

Below: the tracker's headline numbers, what they do and do not prove, the notable cases it lists, and a close look at the one ruling that changes the law this month, the Third Circuit's decision in Thomson Reuters v. Ross.

Quick reference: the tracker's numbers

These come from the tracker page, which does not name its authors and lists only an editor email address. Nothing here has been audited by us, and the site's own wording that 2026 "has already passed" 2025 does not match the equal 86 and 86 counts it shows, a small sign to check dockets yourself.

What do the numbers say about AI legal risk?

Copyright is the center of gravity. The tracker puts 145 cases in the copyright category, well over half of the total, and flags 69 as class actions. The other categories it names are chatbot harm, privacy and biometrics, agentic AI, deepfakes and likeness, antitrust, and AI employment discrimination. The mix matters: copyright suits are about training data and outputs, so they bear on how models get built, while chatbot-harm and agentic-AI cases bear on how products behave.

Two courts dominate. Northern District of California (93 cases) and the Southern District of New York (41) hold more than half the total. That tracks where the labs and the publishers sit. A ruling in either district can set a pattern the others cite.

The most-sued names are the most-exposed. OpenAI is named in 55 cases, nearly one in four. Google, Meta and Microsoft follow, with Anthropic at 18. Being named often is not the same as losing often: only 25 of the 248 are listed as decided and 18 as settled, while 134 are still active. Most of the legal picture is still unwritten.

Plaintiffs are a mix. 118 individuals, 114 companies and institutions, and 16 government enforcers. The government slice is small but matters most for product design, because regulators can ask for changes, not just damages.

Copyright claim stamp, representing the 248 AI lawsuits tracked against 76 companies

Which cases does the tracker highlight?

The page lists several landmarks. We checked the Ross ruling against independent reporting below; the rest are the tracker's summaries and we have not independently verified each date.

Our own coverage of this wave includes the Anthropic $1.5 billion book piracy settlement approval, the follow-up on Anna's Archive and the Bartz fair-use finding, the Sony and Warner suit over song training, and USA Today's $250 million claim against OpenAI.

The ruling that matters this month: Thomson Reuters v. Ross

Of the cases above, the Third Circuit decision is the only one that is both new and appellate. On September 29, 2026 the court affirmed Judge Stephanos Bibas's February 2025 summary judgment for Thomson Reuters, according to LawNext. It held two things:

• Copyrightability. Westlaw headnotes, including short quotes or paraphrases of judicial holdings, are original enough to be protected.
• Fair use. ROSS Intelligence's internal use of the headnotes as training data for its legal search AI was not fair use, even though the tool's outputs did not themselves infringe.

The ruling is reported as the first federal appellate decision on fair use in AI training. LawNext also notes why its reach may be limited: ROSS copied the material before generative AI became widespread, used it to train a non-generative system, and Bibas found the copying was meant to build a product competing directly with Thomson Reuters. Thomson Reuters said it "firmly believes that respecting copyright is essential for fostering innovation," and ROSS has said it will ask the Supreme Court to review.

Hand placing a green pebble into one of two consent bowls, illustrating consent over AI training data

Does Ross decide the OpenAI and Anthropic cases?

No, and the differences are the point. The generative model cases turn on whether training is transformative and whether outputs substitute for the originals. Ross is a competitor-product case about a retrieval tool. Still, an appellate court rejecting an AI-training fair-use defense, even narrowly, gives plaintiffs a citation they did not have last month. Courts in the Northern District of California, which hold the largest share of the tracker's cases, will read the full opinion closely once it is released after redactions.

For the other big pending case, the NYT matter, secondary reports describe cross-motions for summary judgment in early September and briefing running to late November, with no trial date; those reports could not be matched to the docket, so confirm on PACER or CourtListener before relying on the dates.

What should builders do with this?

You are not the defendant, but the cases shape what you can safely use. Practical steps:

1. Know your model's data story. Ask vendors what they have licensed, what settlements they have made, and what indemnity they offer for output claims.
2. Avoid retrieval over content you cannot license. Ross was about copying a proprietary editorial layer to build a rival product; building a competing tool on scraped proprietary summaries is the highest-risk pattern.
3. Log provenance. If you fine-tune, keep records of dataset sources and licenses.
4. Watch the product-liability cases too. Chatbot-harm and agentic-AI categories affect how you design guardrails and human approvals. Our Felony Bench explainer on AI agent legal liability covers the agent side.
5. Teams that deploy agents should add runtime controls. AgentBeam, the agent security platform from the explainx.ai team, stops AI agents before they take dangerous actions, which reduces the odds of becoming a case study.
6. If you work in law, see how legal vendors position themselves in our post on OpenAI Astra for law, given that the Ross dispute began in legal research.

What the tracker does not tell you

It does not weigh cases. A frivolous pro se complaint and a billion-dollar class action each count as one. It does not show outcome odds, and its status buckets are the site's own classification. It is also anonymous, which is not disqualifying for a compilation of public records but means you should treat it as a map to dockets, not a source in itself. Finally, counts depend on what the maintainers chose to include, so a case missing from the list is not proof it does not exist.

Used carefully, it is still a handy index: filter by company to see your vendor's exposure, by category to see where the law is moving, or by court to see which judges you are likely to hear from.

What people are asking

Will AI companies lose most of these cases? The data does not say. With 134 active and only 25 decided, outcomes are open, and the settled count of 18 shows many defendants prefer to pay rather than risk precedent, as in the Anthropic book settlement.

Is training on copyrighted data legal? There is no single answer yet. District courts have split on facts, and the Ross appeal is narrow. Watch the full Third Circuit opinion and any Supreme Court petition.

Where do I follow updates? The tracker, court dockets on CourtListener, and our coverage. We will update this post when the Ross opinion text is released or a new appellate ruling lands.

Related reading

• Anthropic $1.5 billion book piracy settlement approved
• Anna's Archive, book destruction and Bartz v. Anthropic
• Sony and Warner sue Anthropic over songs
• USA Today sues OpenAI for $250 million
• Felony Bench: AI agent legal liability
• OpenAI Astra for law
• AI Lawsuit Tracker

Case counts and statuses are the tracker's figures as of its October 8, 2026 update; this is not legal advice.