AI Copyright Disputes

Who Owns the Brushstroke and the Melody?

AI-Era Copyright Disputes — Images, Music, Characters

2026. 03. 02

TABLE OF CONTENTS

1. My Painting That I Never Painted

(1) Getty Images v. Stability AI: Watermark Reproduction and Trademark Infringement

(2) Andersen v. Stability AI/Midjourney/DeviantArt: Proving Copyright Infringement Through Style Imitation

(3) The Compressed Copy Theory and Its Technical & Legal Issues

2. The Collision Between the Music Industry and AI

(1) RIAA v. Suno: Unauthorized Training on Music Recordings

(2) RIAA v. Udio: The Direct Infringement Liability Debate

(3) Concord Music v. Anthropic: Song Lyrics Output and Licensing Issues

3. Character and Content Copyright

(1) Disney/Universal v. Midjourney: Reproducing Famous Characters and Derivative Work Infringement

(2) Google AI MDL Consolidated Litigation

SECTION 01

1. My Painting That I Never Painted

On November 4, 2025, Justice Joanna Smith of the English High Court delivered a 205-page ruling. The world was watching. The first head-on collision between image-generating AI and copyright. The outcome was unexpected.

Getty Images did not win.

The case began in January 2023. Getty Images filed suit against Stability AI in the English courts. The claim was clear.

However, as the trial progressed, problems emerged. Getty's lawyers could not find evidence that Stable Diffusion's training had taken place within the United Kingdom. The training happened in the United States. UK copyright law applies only to acts committed within the UK. Getty had to withdraw its primary copyright claims.

Two issues remained. First, whether the AI model constituted an "infringing copy." Second, whether trademark infringement had occurred.

Justice Smith answered "no" to the first question. The key passage in her ruling read: "AI model weights are not 'copies' of the images. The model does not store visual information. It merely contains statistically trained parameters."

This was a crucial determination.

Yet Getty did secure a small victory. Trademark. Justice Smith acknowledged that early versions of Stable Diffusion had produced images reproducing Getty's watermark.

Some user-generated images bore the "GETTY IMAGES" watermark. The very watermark embedded in the original photographs. The judge found this constituted trademark infringement, though only "to an extremely limited extent." After Stability AI improved its filtering technology, the watermarks no longer appeared.

The message left by this ruling is complex. For AI companies, it was a sigh of relief. Model weights are not copies. But it was also a warning. If identifiable marks like watermarks appear in outputs, that can constitute trademark infringement.

The fight in the UK came to a temporary close. But a separate lawsuit is proceeding in the US District Court of Delaware. Same parties, same issues, different legal system. What the American court will decide remains unknown.

Sarah Andersen was a webcomic artist. Under the name "Sarah's Scribbles," she was one of the most beloved illustrators on the internet, with millions of followers. In January 2023, she filed a class-action lawsuit along with other artists. The defendants were Stability AI, Midjourney, and DeviantArt. The claim: her art had been used to train AI without authorization.

As of February 2026, this lawsuit is in the discovery phase. Trial is scheduled for September 8, 2026. But important rulings have already been issued.

On August 12, 2024, Judge William Orrick denied most of the defendants' motions to dismiss. He recognized that the plaintiffs' copyright infringement claims deserved to be heard in court.

What caught Judge Orrick's attention was a statement by Stability AI CEO Emad Mostaque. In an interview, Mostaque had said: "We 'compressed' 100,000 gigabytes of images into a 2-gigabyte file. This file can 'reproduce' any of those images."

The judge wrote: "The inference that Stable Diffusion was built in substantial part on copyrighted works, and that its operation necessarily invokes copies or protected elements of those works, is plausible at this stage."

As of January 2026, the lawsuit remains in the discovery phase. Trial is set for September 8, 2026, but an unexpected battle has erupted along the way. A war over expert witnesses.

The plaintiffs put forward Professor Ben Yanbin Zhao as an expert witness. An endowed professor of computer science at the University of Chicago. But this individual had a distinctive background. He was the creator of tools called "Nightshade" and "Glaze."

Nightshade was a kind of "poison." When artists applied the tool to their images, the images appeared unchanged to the human eye, but AI models perceived something entirely different. For example, a human sees a cow in a meadow, while the AI sees a leather handbag on grass. AI models trained on such "poisoned" images would produce increasingly distorted outputs.

The defense attorneys' reaction was immediate. We cannot show our source code and training data to someone who has built a tool designed to sabotage our models.

In June 2025, Judge Lisa Cisneros held a hearing on the matter. She called it "a difficult question." Professor Zhao is an academic researcher, not an employee of a competing company. But his research was adversarial to the defendants' products.

On July 14, 2025, Judge Cisneros issued her ruling. She barred Professor Zhao from accessing the defendants' highly confidential materials. The plaintiffs challenged this decision.

On August 29, 2025, Judge Orrick upheld Judge Cisneros's decision. "Dr. Zhao's research places him in an 'adversarial posture' to the defendants. There is a risk that the defendants' highly confidential information could be inadvertently used, and competitive harm could result."

However, Judge Orrick clarified one point. "Judge Cisneros's ruling does not mean Dr. Zhao cannot testify or assist the plaintiffs as an expert. It only means he cannot view information that the defendants have designated as highly confidential under the protective order in this case."

For the plaintiffs, this was a blow. They argued that Professor Zhao possessed "irreplaceable expertise." But the court found that alternative experts existed. The defense cited Dr. Emily Wenger, who had been publicly identified as an expert in the Google generative AI copyright litigation.

According to the joint status report of October 16, 2025, discovery negotiations between the parties continue. Stability AI agreed to 10 search term strings across 7 custodians. Midjourney consented to 13 search terms across 6 custodians. DeviantArt permitted 8 search terms for 4 custodians.

Midjourney's production of training data has become a separate issue. In July 2025, Midjourney requested an extension of the training data production deadline, which the court granted.

Approximately eight months remain until trial. Both sides are still collecting evidence and preparing experts. But one thing has already become clear. This lawsuit is not simply about whether copyright was infringed. It is also about what "secrecy" means in the age of AI, and what "competition" means.

Meanwhile, around the same time, Midjourney—a key defendant in this case—faced another front. In June 2025, Disney and Universal sued Midjourney for copyright infringement. In September 2025, Warner Bros. joined in. Three of Hollywood's five major studios are now in litigation against Midjourney.

About four years since Sarah Andersen discovered unfamiliar drawings rendered in her artistic style on Twitter. The lawsuit she initiated has become the epicenter of an earthquake shaking the entire AI image generation industry.

In the next section, we examine the "Compressed Copy" theory—one of the key issues in this litigation. Does an AI model truly "compress" and store billions of images?

Internal documents revealing how AI companies collected training data, which images were included, and how copyright protections were circumvented in the process.

The outcome of this litigation will affect the entire US AI industry. If the plaintiffs prevail, the current practice of "training AI on data scraped from the internet" could be fundamentally disrupted.

In the spring of 2024, Judge William Orrick of the Northern District of California was wrestling with a strange question.

An AI model called Stable Diffusion had been "trained" on billions of images. The resulting model file was approximately 4 gigabytes. The total size of the original images was measured in petabytes. Information millions of times larger had been reduced to a fraction of its size. The question was this: Is this 4-gigabyte file a "copy" of the original images?

To answer this question, one must first understand how AI works.

Think of a photo stored on your phone. It is a JPEG file. This file is a sequence of binary digits—zeros and ones. When you open the photo, software decodes these digits and displays the image on screen. The key point is this: opening the same file always produces the same image. One photo, one file. A one-to-one correspondence.

AI models are different. Inside a generative AI model, there are billions of numbers. They are called "weights." These weights contain statistical patterns extracted from millions of training images. Patterns like "cat ears generally follow this curve," or "eyes are typically positioned here." When you type "cat," the model combines these patterns to generate an image that looks like a cat.

But no specific cat photo is "stored" within these weights. It is not a structure where decompressing yields the original, like a JPEG. Even with the same prompt, a different cat image is generated each time. Millions of images condensed into a single model. A many-to-many relationship.

Up to this point, this is the AI companies' argument. They say: Our model is not a copy. We did not "store" images—we "learned" from them. Just as a human painter views thousands of paintings and develops their own style.

But there are uncomfortable facts.

In 2023, researchers entered specific prompts into Stable Diffusion. The model generated images bearing clearly visible Getty Images watermarks. It had reproduced images from the training data almost verbatim. In other experiments, famous photographers' works were restored at the pixel level. The model had "memorized" the originals.

Technical experts call this phenomenon "overfitting" or "memorization." The more frequently an image appears in the training data or the more distinctive its features, the higher the probability the model will reproduce it verbatim. The model does not only learn patterns—it sometimes memorizes the originals themselves.

The legal debate diverges here.

Justice Michael Green of the English High Court ruled in the 2024 Getty Images case: "Model weights themselves are not copies of training images. The weights possess the potential to reproduce images, but they do not contain the images themselves." He distinguished between the potential for reproduction and actual reproduction.

American courts have not yet reached a conclusion. In the Andersen v. Stability AI case, Judge Orrick asked: "If the model is merely a 'tool' as the defendants claim, why can that tool reproduce the plaintiffs' works so accurately?" He did not fully dismiss the plaintiffs' claims. He demanded further evidence.

The key issue is the definition of "reproduction." When the US Copyright Act of 1976 was enacted, the reproduction legislators envisioned was physical reproduction through copiers and printing presses. A world in which original and copy were clearly distinguishable. What now lies before the courts is an entirely different technology. A system that can "reproduce" originals without "storing" them.

Scholars call this the "compressed copy" theory. The claim that an AI model is an extremely compressed form of reproduction of the training data. Unlike a ZIP file, perfect restoration is impossible, but partial restoration is possible. And copyright law recognizes partial reproduction as infringement.

As of February 2026, three federal judges in the US have ruled on the relationship between AI training and copyright. Two leaned in favor of AI companies. One left the door open for copyright holders. But all of these rulings are from the early stages of litigation. They are not final verdicts. Appeals are underway, and discovery is continuing.

SECTION 02

2. The Collision Between the Music Industry and AI

On June 24, 2024, the Recording Industry Association of America (RIAA) filed two lawsuits simultaneously. One against Suno, one against Udio. The plaintiffs were Universal Music Group, Sony Music, and Warner Music Group. All three of the world's major record labels.

Suno is an AI music generation service. When a user types "an upbeat pop song in the 1980s style," the AI creates music to match. Lyrics, melody, arrangement—everything.

The record labels' claims were firm.

The complaint included shocking evidence. When the plaintiffs' investigators entered specific prompts into Suno, the output bore a striking resemblance to Chuck Berry's "Johnny B. Goode" or Jerry Lee Lewis's "Great Balls of Fire." The distinctive rhythms and melodies were reproduced.

Suno's rebuttal was predictable.

In September 2025, the RIAA amended its complaint to add new allegations. Suno had "illegally scraped" recordings from YouTube. Specifically, it had circumvented YouTube's rolling cipher system to download recordings. If true, this would add a violation of Section 1201 of the Digital Millennium Copyright Act (DMCA) to the copyright infringement claims.

But in November 2025, something unexpected happened.

Warner Music Group settled with Suno. The first defection among the three record labels.

The settlement terms were as follows. Suno would phase out its current model. A new platform would launch in 2026. This platform would be trained exclusively on licensed music. Warner's artists could "opt in" to having their music used for training and receive compensation.

The financial terms were not disclosed. But the message was clear. From war to peace, from litigation to partnership.

Universal and Sony's lawsuits continue. But Warner's settlement sent a signal to the other labels. Negotiating at the table may be better than fighting in court.

Udio was practically Suno's twin. An AI music generation service built by former Google DeepMind researchers. Launched in April 2024, it grew rapidly. It received investment from venture capital firm Andreessen Horowitz (a16z), with musician will.i.am also among the investors.

On the very same day it sued Suno, the RIAA also filed against Udio with the same logic. The complaint filed in the Southern District of New York was nearly identical. Unauthorized training, copyright infringement, market substitution.

But the Udio litigation took a different path.

On October 29, 2025, Universal Music Group announced a settlement with Udio. The press release was headlined "Industry-First Strategic Agreement."

The terms were more specific than the Warner-Suno settlement. First, a financial settlement was reached. The amount was not disclosed. Second, a licensing agreement was signed. Udio obtained rights to use Universal's recorded music and publishing catalog. Third, a new service would launch in 2026. This service would operate in a "licensed and protected environment."

Universal CEO Lucian Grainge said: "This agreement demonstrates our commitment to doing right by our artists and songwriters."

Udio CEO Andrew Sanchez was more optimistic: "This moment realizes everything we've built toward. Integrating AI and the music industry in a way that truly champions artists."

Sony Music's lawsuit is still ongoing. But on a front from which Universal and Warner have withdrawn, the dynamics of the battle will inevitably change.

In October 2023, music publishers sued Anthropic. Universal Music Publishing, Concord Music Group, ABKCO. Their claim was this.

The complaint included test results. When the prompt "Tell me the lyrics to Katy Perry's 'Roar'" was entered, Claude output nearly flawless lyrics.

But this lawsuit developed in a different direction from other AI copyright cases.

In January 2025, the parties reached an agreement. Not a settlement ending the lawsuit, but an agreement on the output problem. Anthropic committed to maintaining guardrails preventing Claude from outputting the plaintiffs' lyrics. It also agreed to block the generation of new lyrics.

With this, the "output stage" infringement issue was tentatively resolved. What remained was the "input stage"—whether the mere inclusion of lyrics in training data constitutes infringement.

In March 2025, Judge Eumi Lee denied the plaintiffs' motion for a preliminary injunction. For two reasons. First, the scope of the injunction was too broad. The plaintiffs sought an injunction covering hundreds of thousands of songs, but it was unclear how Anthropic could comply. Second, irreparable harm was not demonstrated. There was no evidence that the existing licensing market had shrunk because of Anthropic.

The next day, the judge also dismissed claims for contributory and vicarious infringement. The reason was that the plaintiffs had failed to prove "direct infringement by a third party." Many of the lyrics-output examples cited in the complaint had been generated by the plaintiffs' own investigators during testing, not by actual users.

However, the direct infringement claim survived. Trial was set for November 18, 2025.

In October 2025, the plaintiffs filed a motion to amend their complaint. They sought to add allegations that Anthropic had downloaded lyrics from pirate sites ("shadow libraries"). This fact had emerged from a separate case, Bartz v. Anthropic.

But Judge Lee denied this motion. The reason: the plaintiffs had not diligently investigated this issue before the close of discovery. Anthropic scored a small victory. But the direct infringement trial still lies ahead.

SECTION 03

3. Character and Content Copyright

On June 11, 2025, Hollywood declared war on AI.

Disney and Universal jointly sued Midjourney. A 110-page complaint. The world's largest entertainment companies had dragged the world's largest image-generation AI company into court.

The complaint's language was blunt. Midjourney is "a bottomless pit of plagiarism" and a "virtual vending machine" that copies and sells Disney and Universal's copyrighted works without authorization.

Evidence photos were placed side by side in the complaint. On the left, images generated by Midjourney. On the right, the original characters. Darth Vader. Elsa. Buzz Lightyear. Shrek. Minions. The Simpsons.

The simple prompt "animated toy" produced Woody and Buzz from Toy Story. The prompt "popular movie screencap" reproduced scenes from specific Disney films. Entering a character's name naturally produced that character.

Disney's chief legal officer Horacio Gutierrez issued a statement: "We are optimistic about the possibilities of AI technology. But piracy is piracy. It is no less infringing when an AI company does it."

According to the complaint, Midjourney had 21 million users and annual revenue of $300 million. Disney and Universal demanded statutory damages of up to $150,000 per infringed work. Over 150 works were listed in the complaint. Potential damages exceed $20 million.

On August 6, 2025, Midjourney filed its answer. It denied all claims.

The defense arguments were as follows. First, fair use. Second, neural networks do not "store" works. They only learn statistical patterns. Third, Midjourney is not liable for user-generated content.

The plaintiffs countered. Midjourney already has technology to filter violent and nude imagery. It could filter copyrighted characters too. It simply chose not to.

This lawsuit is in its early stages. Discovery has not yet begun. But other suits are already following. In September 2025, Warner Bros. also filed a similar lawsuit against Midjourney.

Google is not free from the copyright wars either.

Authors and photographers have sued Google's Bard (now Gemini) and Imagen.

In October 2024, multiple lawsuits were consolidated. The claims are similar to other AI lawsuits. Google scraped data from across the entire internet to train AI without permission. Copyrighted content was included in the process.

In September 2025, the court dismissed claims regarding earlier AI models, but allowed claims regarding Gemini and Imagen to proceed.

In October 2025, the plaintiffs filed for class certification. A hearing is scheduled before Judge Eumi Lee on February 4, 2026.

The key question is this: Can publicly posted content be freely used for AI training, or is permission required?

If Google loses, the damages could be astronomical. After all, the entire internet was the training data. Data deletion or establishment of a licensing framework could be required.

Meanwhile, lawsuits against OpenAI are also being consolidated in the Southern District of New York. On April 3, 2025, the US Judicial Panel on Multidistrict Litigation combined more than 12 lawsuits into a single multidistrict litigation (MDL). The New York Times, the Authors Guild, Raw Story, The Intercept, and the Chicago Tribune are among the plaintiffs. Judge Sidney Stein is presiding.

On October 8, 2025, Judge Stein held a four-hour oral argument on OpenAI's motion to dismiss. A decision has not yet been issued.

A new summary judgment decision on fair use is unlikely before the summer of 2026. Until then, AI companies and content creators must formulate their strategies amid uncertainty.

#AICopyright #StabilityAI #GettyImages #Midjourney #FairUse #CompressedCopy #AITraining #CopyrightLaw #RIAA #Suno #Udio #Anthropic #Claude #SongLyrics #Disney #UniversalMusic #WarnerMusic #OpenAI #GenerativeAI #AIRegulation #ImageGeneration #MusicAI #DeepLearning #ModelWeights #DigitalCopyright #DMCA #ClassAction #AILawsuit #ArtificialIntelligence #CopyrightInfringement

kimkj.com — AI Copyright Disputes Analysis