AI copyright lawsuits are federal cases in which writers, artists, photographers, news publishers, and record labels accuse artificial intelligence companies of copying millions of protected works to train commercial AI systems without permission or payment. The plaintiffs seek damages that could reach $150,000 per work for willful infringement.1Office of the Law Revision Counsel. 17 USC Code 504 – Remedies for Infringement: Damages and Profits The AI developers respond that training a model on existing work is a transformative, noninfringing use. No court has yet issued a definitive ruling on that question, which is why these suits are among the most consequential intellectual property disputes in decades.
What Plaintiffs Say the AI Companies Did
Federal copyright law gives creators the exclusive right to reproduce their works.2Office of the Law Revision Counsel. 17 USC Code 106 – Exclusive Rights in Copyrighted Works When an AI company scrapes a photograph, news article, or illustration and feeds it into a training dataset, it creates a digital copy. Plaintiffs argue that this is mass unauthorized reproduction, and the numbers are large: training datasets for major models contain billions of files, each one a copy of someone’s work ingested without a license.
The developers frame the same act differently. Ingesting data is a technical step in building a new product, they say, not an attempt to exploit the creative value of any individual work. The copies exist to teach statistical patterns. Copyright owners answer that the resulting products are commercial, sometimes generating billions in revenue, and that building them on unlicensed data bypasses the traditional market for creative work.
The exposure is what makes these suits serious. Copyright owners who registered their works before the infringement can elect statutory damages instead of proving actual losses. Those damages run from $750 to $30,000 per work, and up to $150,000 per work if the infringement was willful.1Office of the Law Revision Counsel. 17 USC Code 504 – Remedies for Infringement: Damages and Profits Multiplied across a training set, the potential liability is enormous.
The Fair Use Fight
Fair use is the central battleground. Courts weigh four factors: the purpose and character of the use, the nature of the copyrighted work, how much was used relative to the whole, and the effect on the market for the original.3Office of the Law Revision Counsel. 17 USC 107 – Limitations on Exclusive Rights: Fair Use No single factor decides the outcome.
AI companies lean hardest on the first factor. Training a model, they argue, is a transformative use because it serves a fundamentally different purpose than the original work: a photograph was made to be viewed, and feeding it into a neural network to learn visual patterns is something else entirely. The U.S. Copyright Office, in its 2025 report on AI and copyright, acknowledged that some intermediate copying for nonexpressive purposes can be transformative, but cautioned that using copyrighted works to generate competing content in the same market pushes beyond established fair use boundaries.4U.S. Copyright Office. Fair Use Index
The fourth factor, market harm, may prove decisive. When an AI image generator produces illustrations that compete with the artists whose work trained it, or a language model produces text that substitutes for a journalist’s reporting, courts must weigh whether the AI use displaces sales of the originals. Plaintiffs say the harm is obvious: AI-generated content floods the same markets at lower cost and higher speed.
One early ruling gives a preview. In Thomson Reuters v. Ross Intelligence, a federal court in Delaware granted summary judgment to Thomson Reuters, finding that Ross’s use of copyrighted legal headnotes to build a competing legal research tool was not fair use. The court held that Ross “took the headnotes to make it easier to develop a competing legal research tool,” so the use was not transformative.5United States District Court for the District of Delaware. Thomson Reuters Enterprise Centre GmbH v Ross Intelligence Inc The court was careful to note it was addressing only non-generative AI, but the reasoning about competitive substitution applies directly to arguments in the larger generative AI cases.
When AI Outputs Themselves Infringe
Even if training were found to be fair use, the outputs a model produces can independently infringe. When a user prompts an image generator to create something “in the style of” a specific artist and the result closely mirrors that artist’s unique visual elements, the output may qualify as an unauthorized copy or derivative work. Courts apply a substantial similarity test, asking whether an ordinary observer would recognize that the AI output was taken from a specific copyrighted source.
Copyright protects an artist’s specific expression, not their general style, technique, or method. An AI output that evokes an artist’s aesthetic without copying specific compositional elements sits in legally ambiguous territory, and federal courts remain split on where to draw that line.
Who Pays When Outputs Infringe
Users who prompt AI tools into producing infringing content may be directly liable, but the deeper pockets belong to the companies that built the tools. Vicarious infringement claims target the developer and require proof of two things: the developer had the right and ability to supervise or control the infringing activity, and the developer benefited financially from it.6United States Courts for the Ninth Circuit. 17.20 Secondary Liability – Vicarious Infringement – Elements and Burden of Proof Financial benefit is not in dispute. The fight centers on control. If a developer can filter or restrict infringing outputs and chooses not to, that failure strengthens the plaintiff’s case.
Some AI companies have tried to invoke the DMCA safe harbor, which shields online service providers from liability for user-uploaded infringing content when they meet certain conditions, including terminating repeat infringers and promptly removing infringing material on proper notice.7Office of the Law Revision Counsel. 17 USC 512 – Limitations on Liability Relating to Material Online The safe harbor covers content stored at the direction of users, not the platform’s own actions. An AI company that copies works into its own training dataset is doing the copying itself, which makes safe harbor a poor fit for training-related claims.
The DMCA Metadata Claim
A separate line of attack involves copyright management information. Federal law prohibits intentionally removing or altering identifying information attached to a copyrighted work when the person knows or should know that doing so will facilitate infringement.8Office of the Law Revision Counsel. 17 USC Code 1202 – Integrity of Copyright Management Information That information includes watermarks, author names, and licensing terms embedded in digital files.
Plaintiffs argue that AI training pipelines systematically strip this data. When a model ingests millions of images, metadata like photographer credits and licensing terms is typically discarded. If the model then produces outputs resembling those works, the original creators have no way to trace how their work was used. AI developers respond that the removal is an automatic byproduct of how neural networks process data, not an intentional act of concealment. Courts are evaluating whether automated stripping at industrial scale meets the statute’s knowledge requirement.
Damages under this provision are separate from ordinary infringement damages. Courts can award between $2,500 and $25,000 per violation.9Office of the Law Revision Counsel. 17 USC 1203 – Civil Remedies Because each stripped file can count as a separate violation, the aggregate exposure adds another layer of financial risk on top of standard infringement damages.
Where the Major Cases Stand
Several high-profile cases are moving through federal courts, each representing a different creative industry. None has reached trial yet, but rulings on motions to dismiss and discovery disputes are beginning to shape the ground.
The New York Times v. Microsoft and OpenAI
Filed in December 2023, this case alleges that OpenAI and Microsoft used millions of Times articles to train large language models without permission.10The New York Times Company. The New York Times Company v Microsoft Corporation, OpenAI, Inc, et al – Complaint The Times argues those models can reproduce near-verbatim excerpts of its reporting, directly competing with its subscription business. The case remains in active litigation before Judge Sidney H. Stein in the Southern District of New York, with filings as recent as May 2026.11CourtListener. The New York Times Company v Microsoft Corporation No trial date has been set publicly.
Andersen v. Stability AI
A group of visual artists filed this class action alleging that Stability AI, Midjourney, and DeviantArt used billions of scraped images to train image-generation models without consent or compensation.12Justia. Andersen et al v Stability AI Ltd et al In August 2024, the court denied the defendants’ motion to dismiss claims for direct and induced copyright infringement, finding them plausible. The case is now in discovery, with a trial date set for September 8, 2026.
Getty Images v. Stability AI
Getty Images sued Stability AI over the alleged scraping of millions of stock photographs and their associated metadata. Some AI-generated outputs appeared to reproduce a distorted version of the Getty Images watermark, raising both copyright and trademark claims.13CourtListener. Getty Images (US) Inc v Stability AI Ltd The U.S. case, now in the Northern District of California, survived a partial motion to dismiss in April 2026, with trial scheduled for January 2028. A parallel case in the United Kingdom has already gone to trial, with a judgment issued in late 2025.
The Music Industry Cases
The recording industry opened a separate front in 2024 when major labels including Sony Music, UMG Recordings, and Warner Records sued AI music generators Suno and Udio, alleging the services were trained on decades of copyrighted sound recordings to produce output that imitates genuine recordings. Music publishers have also sued Anthropic, claiming its Claude chatbot reproduces copyrighted song lyrics. In the Anthropic case, the court denied a motion to dismiss claims for contributory infringement, vicarious infringement, and removal of copyright management information, allowing the case to proceed.
Licensing Deals as an Off-Ramp
Not every interaction between AI companies and content owners ends in court. Some companies have opted to license training data. Shutterstock signed a six-year agreement with OpenAI providing access to its image, video, and music libraries for AI training.14Shutterstock. Shutterstock Expands Partnership with OpenAI, Signs New Six-Year Agreement to Provide High-Quality Training Data Shutterstock compensates its contributors through a dedicated fund that pays based on the role each contributor’s work played in training and includes ongoing royalties tied to licensing activity.
Plaintiffs in the pending suits point to deals like this as proof that a licensing market exists and that defendants chose to skip it. AI companies describe voluntary licensing as a business decision rather than an admission that training without a license is illegal. The courts will decide which view prevails.
What Congress Is Considering
Congress has begun considering legislation that would address AI and copyright directly, though no bill has become law. The 119th Congress (2025–2026) has seen multiple proposals, including the TRAIN Act (Transparency and Responsibility for Artificial Intelligence Networks Act) introduced in both chambers, the NO FAKES Act of 2025 addressing AI-generated likenesses, and the Copyright Labeling and Ethical AI Reporting Act.15U.S. Copyright Office. Legislative Developments Most focus on transparency, such as requiring AI companies to disclose when copyrighted works appear in their training data.
Until a bill passes, the rules will be set almost entirely by the courts. The cases now in litigation are likely to produce the first binding precedents on whether large-scale AI training qualifies as fair use, and those rulings will determine whether the industry has to negotiate licenses or can keep building on freely scraped data.