AI and Copyright: The Legal Fight Reshaping Who Owns Creative Work

Artificial intelligence has never been more useful, or more legally exposed. Over the past year, courts on two continents have started answering the question every publisher, musician, and AI company has been asking since ChatGPT and Midjourney went mainstream: who owns the work that trains, and the work that comes out of, an AI system?

A U.S. court has now approved the largest copyright settlement in American history against an AI company. A German court has ruled that training an AI model in the United States can still get you sued in Europe. And a federal appeals court has, for the first time, been asked directly whether training AI on copyrighted material counts as fair use. None of this is theoretical anymore — it is actively deciding how much AI companies pay, what they can legally build, and what protection creators have.

Two Questions Behind Every AI Copyright Case

Nearly every lawsuit in this space comes down to two separate issues. First: is it legal to train an AI model on copyrighted work without permission? Second: once AI produces something, can anyone copyright it? Courts are answering these two questions very differently and mixing them up is the most common mistake in how this topic gets reported.

The $1.5 Billion Settlement That Got Everyone's Attention

On July 20, 2026, a federal judge in California gave final approval to a $1.5 billion settlement in Bartz v. Anthropic, the largest copyright class-action recovery in U.S. history. The case centered on authors and publishers whose books Anthropic had downloaded from pirate sites to train its Claude models. The payout covers roughly 480,000 to 500,000 works, at just over $3,000 per book, and Anthropic must also destroy the pirated files and any copies derived from them.

Here is the part most coverage leaves out: the settlement never actually decided whether training AI on copyrighted text is fair use. Anthropic paid to make a piracy problem disappear before a judge could rule on the bigger legal question. That question is still open, and it is being fought out right now in other courtrooms.

Fair Use: The Fight Nobody Has Actually Won Yet

Two live cases are pushing that exact question toward a real answer. In Thomson Reuters v. ROSS Intelligence, a court already ruled that training an AI legal-research tool on Westlaw's copyrighted headnotes was not fair use. ROSS appealed, and on June 11, 2026, the case became the first AI-training dispute to reach oral argument before a U.S. Court of Appeals, the Third Circuit is now expected to hand down the first real appellate precedent on this exact issue.

Meanwhile, in Sony Music v. Suno, discovery reportedly turned up tens of thousands of copyrighted recordings inside Suno's training data. Universal Music Group and Warner Music Group already settled with the rival platforms Udio and Suno rather than risk a ruling. Sony didn't, and its case reached summary judgment in July 2026,  with Suno's entire defense now resting on a single argument: fair use.

Europe Just Closed the Loophole Everyone Was Counting On

The most important ruling of the summer may be the one AI companies saw coming the least. On July 31, 2026, a Munich court ruled against Suno in a case brought by the German rights society GEMA, the first European court decision on a generative AI music tool. GEMA showed the court that Suno's outputs closely resembled six well-known songs, including "Rasputin" and "Mambo No. 5."

The court ordered Suno to stop training on the works, stop making the model available, and pay damages. But the real headline is jurisdictional: Suno's training happened entirely on U.S. servers, and the German court asserted jurisdiction anyway, applied U.S. copyright law, and rejected the fair-use defense. Training your model somewhere with looser rules no longer guarantees you are safe once that model reaches users elsewhere.

Can Anyone Actually Copyright What AI Makes?

Separate from the training-data fight, courts have been far more consistent on this second question. In Thaler v. Perlmutter, a computer scientist tried to copyright an image his AI system generated, listing the AI itself as author. The D.C. Circuit ruled in March 2025 that human authorship is a bedrock requirement of copyright law,  the Supreme Court declined to revisit that ruling in March 2026, leaving it as settled law for now.

The U.S. Copyright Office backs this up: AI-generated material can be copyrighted only where a human contributed real creative expression,  editing, arranging, meaningfully shaping the result. A prompt alone, no matter how detailed, does not make you an author.

What This Actually Means Going Forward

For AI companies, licensing before you train is becoming the safer bet, the settlements piling up in the music industry are pushing the entire sector in that direction. For publishers, musicians, and artists, both the Anthropic and GEMA cases prove there is real financial leverage in pursuing these claims, whether through U.S. class actions or European collective-licensing bodies. And for any business publishing AI-assisted content commercially, keeping a record of genuine human creative input is now the single most practical way to protect it.