AI-Training Lawsuits by Creators Undermine Copyright Law, Policy

Copyright creators, including pop-culture giants Walt Disney Co. and Universal Studios Inc., recently sued several artificial intelligence media-generators for allegedly profiteering off of 100 years of copyrighted works—to the tune of hundreds of millions of dollars.

Some parts of the suits may have merit. Copying is copying. But an expansionist theory asserted by the copyright creators—about what material AI models may train on—could undermine core copyright principles aimed at expanding, not constricting, creative expression.

Worse yet, the theory threatens to cut off burgeoning AI media-generators while costing copyright creators a unique chance to cash in on the AI-user-generated creative boom—a lose-lose all around.

Plaintiffs’ Perspective

Principally, the copyright creators claim that the AI media-generators are committing double-layered copyright infringement: illegally copying their works at the input/learning stage and then creating or letting users create unauthorized derivative works at the output/generation stage.

Disney and Universal pull no punches in their lawsuit. They aggressively brand one AI media-generator as “the quintessential copyright free-rider and a bottomless pit of plagiarism.”

As to the AI training stage, the copyright creators claim that scraping, downloading, and reformatting vast amounts of images and text data from the web—needed to train generative text-to-image AI models that in turn generate hyper-realistic images from prompts—literally violates the “copy” in copyright law.

Copyright Concerns

The copyright creators are literally correct. It takes little to violate copyright holders’ reproduction rights. A single unauthorized copy is enough. And unlike some foreign copyright frameworks, the US doesn’t exempt private use from technical copyright infringement, even if the copy isn’t shared with others.

But not all copying is illegal. Training AI on copyrighted works may constitute fair use—copyright infringers’ multi-factor defense du jour.

Using copyrighted works to train innovative AI media-generator models is arguably highly transformative: The purpose is to create something new rather than supplant the original works.

AI training (and even certain outputs) therefore aren’t legitimate market substitutes for copyright creators’ intellectual property. For example, is a fan really going to skip the next Marvel Cinematic Universe installment due to an AI-generated image of Iron Man on a beach sipping a mai tai?

Policy concerns exist, too. Copyright law is meant to fulfill the constitutional mandate to “promote the Progress of Science and useful Arts.” AI is another tool for creators, which, if unduly limited, will stifle creative expression.

And just because AI models train on copyrighted works doesn’t mean they always generate copyright-infringing works. All artistic works, whether created by humans or AI, draw inspiration from some earlier work. Inspiration is different from infringement.

Courts confronted with these AI-training copyright concerns in the context of training AI models on copyrighted literary works so far have sided with AI, including recent decisions in cases involving Meta Platforms Inc.'s and Anthropic PBC’s large language models.

Threat to AI

If copyright creators succeed in their lawsuits, the profitability of AI media-generators could crater.

Determining copyright ownership, locating copyright owners, negotiating copyright licenses, managing license payments/royalties, and establishing system protocols to identify and filter out unlicensed copyrighted works for the countless images across the web is logistically impractical and financially unfeasible.

Even if it could be done, consumer interest in AI media-generators likely would wane—leading to less consumer revenue—due to creative limitations required by the need to filter out unlicensed copyrighted works, the overall reduced performance or bias of the AI media-generator models themselves, or both.

All this would disproportionately affect smaller, growing AI developers that lack the resources to solve these copyright conundrums at scale, stifling innovation and fostering a monopolistic environment where only the biggest AI players can compete.

Disney and Universal are adamant that AI media-generators could stop their infringement easily. But they don’t argue that the AI media-generators could stop it cheaply or do it and survive.

Missed Opportunity

The AI-media-generation piracy dispute could be an opportunity to grow the pie for all, if it isn’t viewed through a zero-sum lens.

Cues could be co-opted from the music industry, which experienced an existential piracy threat in the 2000s. Display-rights organizations could be created to track AI-generated content and monetize it for copyright creators through blanket licenses, like performance-rights organizations do for songwriters whose songs are performed publicly outside their control.

An act of Congress would be required. But this AI-training-copyright dispute needs a long-term solution; AI isn’t going anywhere.

The cases are Disney Enterprises, Inc. v. Midjourney, Inc., C.D. Cal., No. 2:25-cv-05275, complaint filed 6/11/25; Bartz v. Anthropic PBC, N.D. Cal., No. 3:24-cv-05417, decided 6/23/25; and Kadrey v. Meta Platforms, Inc, N.D. Cal., No. 3:23-cv-03417, decided 6/25/25.

This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law, Bloomberg Tax, and Bloomberg Government, or its owners.

Author Information

Nicholas J. Schneider is a member (partner) with Eckert Seamans’ intellectual property litigation and commercial litigation groups.

Write for Us: Author Guidelines