In 2025, the number of AI-related copyright lawsuits filed in the United States grew from roughly 30 to over 70, according to the Copyright Alliance. That is close to a tripling in a single year. A year ago, most disputes over AI training data and AI-generated output were being settled quietly, away from public view. Now courts are ruling on the merits, a jury trial is opening this month on some of the same underlying questions, and regulators outside the US have started writing rules of their own. Here is what actually changed, and why it now matters even for people who have never trained a model.
Why the numbers are climbing
Three things are driving the jump. First, far more companies are training models on large bodies of text, images and video than were doing so even two years ago, and far more of them are shipping products that generate content for paying customers. Every one of those pipelines touches copyrighted material somewhere, whether in the training set or in what the model outputs. Second, provenance disputes are multiplying. A lawsuit used to ask one question: did you copy my work? Now it can ask several at once: who owns the material a model was trained on, who owns what the model produces, and whether a license for one covers the other. Third, courts are less willing to let these questions get resolved behind closed doors. Early AI copyright disputes mostly ended in private settlements with no public ruling attached. That is changing. Judges are now issuing decisions that other plaintiffs and defendants can point to, which in turn encourages more filings rather than fewer.
Two cases that show the pattern
Bartz v. Anthropic is one clear data point. The case reached a $1.5 billion settlement, with final court approval on 20 July 2026, one of the largest copyright settlements on record and a sign that the financial stakes in these disputes are no longer hypothetical.
Andersen v. Stability AI is a different kind of milestone. A jury trial in the case is opening this month, and it is described as the first US trial to reach a jury verdict on AI image generation and copyright. No verdict exists yet, and nothing here should be read as predicting one. What matters right now is that the case has moved past settlement talks and motions, into a courtroom where a jury, not a private negotiation, will hear the arguments. That alone marks a shift from where this area of law stood twelve months ago.
What regulators are doing about it
Litigation is not the only place these questions are being addressed. The EU AI Act's Article 50 transparency obligations came into force on 2 August 2026, requiring providers and deployers to disclose when content is AI-generated or AI-manipulated. Non-compliance can carry fines of up to EUR 15 million or 3 percent of global annual turnover, whichever is higher. That is a separate mechanism from copyright litigation, but it addresses a related problem: if nobody can tell what is AI-generated, provenance disputes get harder to untangle everywhere, not just in court. Anyone building or publishing AI-generated content into the EU market now has a concrete disclosure obligation to meet, not just a risk to watch from a distance.
What this means for you
None of this means every company using AI is about to be sued. It means the questions these cases are asking, about who owns training data and who owns generated output, are no longer abstract. They are being decided in real courtrooms with real financial consequences, while regulators move in parallel to require more disclosure around AI-generated content. If your business creates, publishes or distributes AI-generated material and needs to understand what the EU AI Act's transparency rules actually require, the compliance picture is laid out in our EU AI Act guide.
