case-law

Recent AI Copyright Cases: What's Happened So Far in 2026

By Nora Whitfield · August 27, 2026

Recent AI Copyright Cases: What's Happened So Far in 2026

A note before we get into it: this is a pulse-check on active litigation, not a permanent reference. AI copyright law is moving fast enough that a post like this needs periodic updates just to stay accurate, and the state of play described here reflects where things stood as of this article's publish date. If you're reading this months later, treat the broad patterns as durable and double-check specific case statuses. For the encyclopedic, continuously maintained rundown of every major suit, see our complete guide to every AI copyright lawsuit worth knowing. This post is the news version. That one's the reference desk.

So, what's actually happened lately?

The Thaler saga is finally, fully over

The longest-running authorship case in this space closed for good. Stephen Thaler spent years trying to register a copyright with his AI system listed as the sole author. He lost at the Copyright Office, lost in district court, lost at the D.C. Circuit in March 2025, and on March 2, 2026, the Supreme Court denied certiorari, letting the appellate ruling stand as final. There's no more appeal left to file. The rule is locked in: a work needs a human author, full stop, and no future case is going to relitigate that particular question. Read the details in our piece on Thaler v. Perlmutter.

What's interesting is how little practical difference this made day to day. Thaler was always an edge case, someone deliberately trying to remove the human from the equation to force a clean test. The far more common, and far messier, disputes are happening elsewhere.

The training-data fair use fight keeps splitting down the middle

If there's one throughline across 2026's biggest rulings, it's that judges are refusing to treat "AI training" as one uniform legal category. The pattern that's emerged: training on data you legally acquired tends to get more sympathetic fair use treatment than training on pirated material, and courts are willing to slice a single case into pieces that come out different ways.

The Anthropic settlement is the clearest example. A judge found that training on lawfully purchased and scanned books could qualify as fair use, a genuinely favorable ruling for AI developers, but drew a hard line at Anthropic's use of pirated book copies to build its training library. That combination, partial fair use win paired with a piracy loss, led to a $1.5 billion settlement, the largest copyright recovery figure this fight has produced.

Meanwhile the New York Times' case against OpenAI and Microsoft remains one of the most closely watched suits still working through the system, with the added wrinkle of claims about direct output reproduction, not just training. Getty Images' parallel suits against Stability AI in the UK and US are producing different signals in each jurisdiction, a reminder that this fight isn't going to resolve into one global rule even once individual cases wrap up.

Music publishers went after Suno and Udio, and it's a different kind of case

Where the book and news publisher suits are mostly about text generation, the music industry's fight with AI song generators Suno and Udio centers on something more direct: allegations that these tools were trained on copyrighted sound recordings without licenses, and that their output can closely mirror existing songs. This is a live, evolving front, and one worth watching separately from the text-and-image cases because the technical and licensing questions in music are genuinely different. Full breakdown in the Suno and Udio lawsuits, explained.

Character and image generation is drawing its own lawsuits

Disney and Warner Bros. brought suit against Midjourney specifically over its ability to generate images of their copyrighted characters on request. This is a notable branch off the main training-data fight, because the claim isn't really about the abstract legality of training on a broad image dataset. It's about a tool that, by design, can output recognizable protected characters when asked, which the studios argue is a different and more direct kind of infringement than a model that merely absorbed general artistic style.

The Copyright Office's own turmoil became part of the story

Separately from the courts, the Copyright Office itself has had a rocky stretch. The pre-publication version of Part 3 of its AI report, covering how training on copyrighted works should be handled, was released on May 9, 2025, less than 24 hours before the dismissal of the sitting Register of Copyrights, Shira Perlmutter. Whether the timing was related to the report's content, which leaned skeptical of broad fair use claims for AI training, has never been fully clarified, but the episode added a layer of political uncertainty to guidance that's supposed to be a stable reference point. We walk through the full three-report series in the US Copyright Office's 3-part AI report, explained.

What to watch next

A few threads are worth tracking as this keeps developing. Whether the "purchased data is more defensible than pirated data" line from the Anthropic settlement becomes a durable rule other courts adopt, or whether it stays specific to that case's facts. Whether output-similarity claims, like the ones against Midjourney, start appearing more often as a cleaner, more provable alternative to sprawling training-data suits. And whether Congress moves on any of the AI transparency and disclosure bills that keep getting introduced and stalling, which we cover separately in our piece on the Generative AI Copyright Disclosure Act.

None of this is settling into a single clean rule anytime soon. If anything, the pattern for 2026 has been the opposite: more granularity, more case-specific splitting, and less appetite from courts to hand down one sweeping answer that covers the whole category of "AI and copyright" at once.