Everything You Need to Know About the AI Copyright Cases Working Their Way Through the Courts

There are currently 23 active copyright lawsuits in US federal courts involving AI companies as defendants. The cases span training data, output similarity, and the right of publicity. None of them have produced controlling precedent. All of them are being watched by everyone in the AI industry with the intensity of people waiting for a verdict that will restructure the economics of a sector.

The Training Data Cases

The foundational question: does training an AI model on copyrighted works without licence constitute copyright infringement?

The AI companies’ position: training is transformative use under the fair use doctrine. The model learns from the works but does not reproduce them; it produces outputs that are not themselves copies of the training works.

The plaintiffs’ position: the copying involved in training is substantial, systematic, and commercial. Fair use requires a case-by-case analysis that cannot be resolved with a blanket “training is fair use” rule.

The current state: the New York Times case against OpenAI is the furthest along and has survived a motion to dismiss on both the direct infringement and the vicarious liability claims. The judge’s ruling on the motion was careful to be narrow — surviving a motion to dismiss does not mean the claims will succeed at trial — but the survival of the case means it will proceed to discovery, which will produce documents that will be consequential regardless of the trial outcome.

The Output Cases

A separate class of cases concerns AI outputs that are alleged to be substantially similar to specific copyrighted works. These cases are harder for plaintiffs in some ways — substantial similarity requires showing the specific output resembles the specific work, which requires the AI to have essentially memorised the work — and potentially more dangerous for AI companies in others, because memorisation of training data is a documented phenomenon rather than a theoretical concern.

What a Resolution Looks Like

The most likely resolution is not a single landmark ruling but a pattern of settlements that establish de facto licensing norms. OpenAI has already licensed content from The Associated Press, News Corp, and several other major publishers. The economic logic of licensing, rather than litigation, is compelling for both sides: publishers get recurring revenue, AI companies get legal certainty. The holdouts — organisations that believe their claims are strong enough to produce a ruling that establishes broader principles — are betting that a court victory would be worth more than a licensing deal. Some of them will be right.