Training Data Lawsuit News: Anthropic, OpenAI, and Fair Use

The wave of AI training data copyright lawsuits has produced its first major rulings, a record $1.5 billion settlement, and a growing split among federal judges over whether using copyrighted books, articles, images, and music to train machine-learning models counts as fair use. As of mid-2026, two California judges have sided with AI developers, a Delaware court has sided with the copyright owner, an appeal that could set the first binding precedent is before the Third Circuit, and roughly 50 active cases remain pending across multiple courts.

How the Fair Use Question Has Split the Courts

Nearly every one of these cases turns on the same question: does feeding copyrighted works into a model qualify as fair use? Courts weigh four factors under U.S. copyright law — purpose and character of the use, nature of the work, amount copied, and effect on the market for the original. Three district court rulings in 2025 reached notably different conclusions.

Anthropic: Training Called “Spectacularly” Transformative

On June 23, 2025, Judge William Alsup of the Northern District of California ruled in Bartz v. Anthropic that using copyrighted books to train large language models is fair use. He called the process “quintessentially transformative,” comparing it to a human reader absorbing ideas rather than replacing the originals.¹ On market harm, Judge Alsup found no evidence that Anthropic’s Claude chatbot produced exact copies of protected works, and he rejected the argument that AI-generated competition could count as market injury, likening it to “training schoolchildren to write well.”²

Judge Alsup drew a sharp line between training and hoarding, however. He held that Anthropic’s storage of more than seven million books downloaded from the pirate sites LibGen and PiLiMi was not fair use, calling it a “use — and not a transformative one” that would go to trial.¹

Meta: Fair Use Even With Pirated Sources

Two days later, Judge Vince Chhabria of the same court ruled in Kadrey v. Meta Platforms that Meta’s use of copyrighted books to train its Llama models was fair use, and he went further than Judge Alsup by holding that obtaining the books from pirate shadow libraries did not defeat the defense. He reasoned that how a copy was obtained is generally irrelevant when the ultimate purpose is highly transformative and the output does not substitute for the original.³

Judge Chhabria also introduced a “market dilution” theory. He acknowledged that large language models can generate “millions of secondary works” in a fraction of the time a human could, potentially flooding creative markets, and signaled that this kind of indirect market harm could tip the balance against a developer.⁴ But the authors presented “no meaningful evidence on market dilution at all,” so Meta won on the record before the court.³ He emphasized that the ruling was narrow, binding only the thirteen named plaintiffs and not declaring Meta’s practices generally lawful. Claims that Meta “seeded” pirated books back onto torrent networks remain active.⁵

Thomson Reuters: Training Is Not Fair Use

The third ruling cut the other way. On February 11, 2025, Judge Stephanos Bibas, sitting by designation in the District of Delaware, granted summary judgment to Thomson Reuters, finding that Ross Intelligence infringed 2,243 Westlaw headnotes used to train a competing legal search tool and that the use was not fair.⁶ The decision reversed the same judge’s 2023 order, which had sent most issues to a jury. Judge Bibas said he had previously placed too much weight on overlap between the headnotes and underlying judicial opinions, and now recognized that Thomson Reuters’s editorial discretion was enough to clear the low bar for copyright protection.⁷

On fair use, he found Ross’s copying commercial and not transformative because the company built a direct market competitor to Westlaw. He distinguished the case from software-related precedents like Google v. Oracle, noting that Ross’s copying of written text was not “reasonably necessary” for interoperability or any new function.⁶ On market harm, the court held that the potential market for AI training data licensing was a relevant derivative market that Ross threatened. Judge Bibas explicitly noted that only “non-generative AI” was before him and cautioned against reading the ruling too broadly.⁸

The $1.5 Billion Anthropic Settlement

The pirated-books claim in Bartz v. Anthropic never reached trial. In September 2025, the parties announced a $1.5 billion class action settlement, the largest payout tied to an AI copyright dispute to date.⁹ Anthropic agreed to pay a minimum of roughly $3,000 per eligible work to copyright owners whose books appeared in the LibGen or PiLiMi datasets the company had downloaded.¹⁰ The settlement fund is non-reversionary; leftover money goes to class members rather than back to Anthropic.¹¹

To qualify, a work must have an ISBN or ASIN and a timely U.S. copyright registration. The finalized list contains 482,460 eligible works.¹¹ Anthropic also agreed to destroy all pirated books and any copies.¹² The deal includes no admission of liability, and final approval is not yet granted. Judge Alsup retired at the end of 2025, and the case is now before Judge Martinez-Olguin, with a fairness hearing scheduled for May 14, 2026.

The New York Times Case Against OpenAI

The highest-profile active case is The New York Times Co. v. Microsoft Corp. and OpenAI, filed in December 2023 in the Southern District of New York. The Times alleges that OpenAI and Microsoft used millions of its articles to train generative AI models and that the resulting tools compete directly with its publishing business. The paper is seeking billions of dollars in statutory and actual damages, a permanent injunction, and the destruction of models trained on its content.¹³

On March 26, 2025, Judge Sidney Stein denied OpenAI’s motion to dismiss the direct and contributory infringement claims, while dismissing some DMCA and unfair-competition claims with leave to amend. Discovery has produced orders compelling OpenAI to preserve and produce massive volumes of ChatGPT output logs, including a November 2025 order for 20 million de-identified logs and March 2026 orders for an additional 78 million and 10 million. Summary judgment briefing concluded in April 2026, and a ruling is expected in the third quarter of 2026.¹³

Consolidated Author Lawsuits Against OpenAI

The Times case is one of twelve lawsuits against OpenAI and Microsoft that the U.S. Judicial Panel on Multidistrict Litigation consolidated in April 2025 in the Southern District of New York as In re OpenAI, Inc. Copyright Infringement Litigation. Plaintiffs include authors Ta-Nehisi Coates, Michael Chabon, Junot Díaz, Sarah Silverman, John Grisham, George Saunders, Jonathan Franzen, and Jodi Picoult, along with news organizations including the Daily News and the Center for Investigative Reporting.¹⁴

In October 2025, the court denied a motion to dismiss the consolidated action, finding that some plaintiffs had sufficiently alleged their works appeared in AI outputs.⁵ Discovery is underway. New complaints keep coming: on June 30, 2025, another group of authors filed Denial v. OpenAI in the Northern District of California, alleging training on books from shadow libraries.¹⁵

Image Generators: Stability AI, Midjourney, and the Studios

Visual artists filed some of the earliest cases. Andersen v. Stability AI, brought by artist Sarah Andersen and others against Stability AI, Midjourney, DeviantArt, and Runway AI, has survived multiple motions to dismiss and is now on its third amended complaint, filed in February 2026. A summary judgment hearing is scheduled for February 2027 and trial for April 2027.¹⁶¹⁷

Getty Images has pursued Stability AI on two continents. In the United Kingdom, a November 2025 High Court judgment found that Stability AI’s inclusion of Getty trademarks in AI-generated images constituted trademark infringement and that the model provider, not the user, bears responsibility. The UK court also confirmed that Getty’s works were used to train Stable Diffusion and held that intangible articles like AI models can be subject to copyright infringement claims.¹⁸ Getty’s U.S. case, now in the Northern District of California, has a jury trial set for January 2028, with mediation ordered by October 2026.¹⁹

Major studios have also joined in. In June 2025, Disney, NBC Universal, and DreamWorks filed a 110-page complaint against Midjourney in the Central District of California, calling the image generator a “bottomless pit of plagiarism” that reproduces characters like Yoda and Marvel heroes without internal safeguards. The studios are seeking an injunction that could force Midjourney to implement copyright filters or temporarily shut down.²⁰

The Music Industry Track

Music copyright holders are running a different play. Universal, Sony, and Warner sued the AI music generators Suno and Udio in 2024, coordinated by the RIAA, alleging mass infringement. Those cases produced settlements rather than trial rulings. Universal settled with Udio in October 2025, and Warner settled with both Udio and Suno in November 2025.²¹ The Universal–Udio deal includes licensing agreements for a revamped platform launching in 2026 that will use AI trained only on authorized music, with artists opting in and being paid for both training and outputs.²² Sony has not settled with either company.

Separately, music publishers Universal, Concord, and ABKCO are suing Anthropic over Claude’s alleged reproduction of copyrighted song lyrics. In March 2026, the publishers filed a motion asking the court to rule before trial that Anthropic infringed and to reject the fair use defense, arguing their case is distinguishable from the book-author litigation because they have “overwhelming” evidence of direct lyric reproduction.²³ The court previously denied Anthropic’s motion to dismiss claims for contributory infringement, vicarious infringement, and removal of copyright management information.²⁴

The settlements themselves have spawned new litigation. In June 2026, the American Federation of Musicians sued Universal and Warner in the Southern District of New York, alleging the labels licensed member recordings to Suno and Udio without compensating the performing musicians, in violation of the union’s collective bargaining agreement.²¹

The First Appellate Ruling Is Coming

The first appellate review of a fair use ruling in an AI training case is now underway. The Third Circuit granted Ross Intelligence’s petition for review of the Thomson Reuters ruling, making it the first federal appeals court to take up an AI copyright case.²⁵ Ross filed its opening brief in September 2025, and oral argument took place on June 11, 2026. The case has drawn amicus briefs from the Electronic Frontier Foundation, the Internet Archive, the American Library Association, Public Knowledge, multiple AI companies, and groups of copyright professors.²⁶ A ruling could establish the first binding appellate precedent on fair use and AI training.

The Ninth Circuit is separately weighing Doe v. GitHub, over Microsoft’s Copilot coding assistant. That appeal, which addresses whether the DMCA requires “identical” copies for liability, had oral argument on February 11, 2026, and a decision is pending.²⁷

The Parallel Licensing Market

While the lawsuits move forward, a market for licensed training data has grown quickly. Throughout 2025, OpenAI signed content deals with the Associated Press, Axios, the Guardian, the Washington Post, and Schibsted Media, among others. Google partnered with the AP for real-time news in its Gemini chatbot and launched an AI pilot with publishers including Der Spiegel, El País, and the Washington Post. In December 2025, Meta signed multi-year licensing agreements with seven publishers, including CNN, Fox News, and USA Today, for content to feed its Llama model.²⁸

The New York Times has taken a split approach. It signed a deal with Amazon in May 2025 to license stories and recipes for Alexa and proprietary AI models, even as it continues to litigate aggressively against OpenAI.²⁸ In music, Spotify has established AI licensing deals with Sony, Universal, and Warner.²⁹ A startup called Prorata launched a revenue-sharing model for its AI search engine; by June 2025, more than 500 publishers had signed on.

Legislation on Both Sides of the Atlantic

On February 10, 2026, Senators Adam Schiff and John Curtis introduced the Copyright Labeling and Ethical AI Reporting (CLEAR) Act, which would require AI developers to submit a detailed summary of every copyrighted work used in training datasets to the U.S. Copyright Office at least 30 days before commercially releasing a model. The bill would create a public database of those disclosures and allow copyright owners to sue developers who fail to provide notice, with penalties of $5,000 per instance and a $2.5 million cap on total civil penalties.³⁰

In the European Union, the framework is further developed. The EU Copyright Directive permits text and data mining for any purpose unless rights holders have expressly opted out, and the EU AI Act requires providers of general-purpose AI models to honor those opt-outs and publish a “sufficiently detailed summary” of training content, regardless of where the model was trained. A review of the Copyright Directive is legally scheduled for June 2026 and may address ongoing ambiguities about what counts as a valid opt-out and whether AI outputs containing substantial portions of protected works trigger separate infringement liability.³¹

Where Things Stand

The picture as of mid-2026 is one of active disagreement. Two Northern District of California judges have found AI training transformative and fair, while a Delaware court reached the opposite conclusion for a non-generative tool, and those judges disagree with each other on whether pirated source material matters. The Third Circuit appeal of the Thomson Reuters ruling could produce the first appellate precedent to resolve at least some of that tension. The OpenAI MDL and the Times case are heading toward potential summary judgment rulings that will test fair use on a much larger factual record. With roughly 50 active copyright lawsuits between AI companies and the entertainment industry alone,²⁹ and billions of dollars at stake in both litigation and licensing revenue, the legal boundaries of AI training remain among the most consequential open questions in American copyright law.

1
Goodwin. District Court Issues AI Fair Use Decision
2
Reed Smith. A New Look at Fair Use in AI Copyright Training
3
Justia. Kadrey et al v. Meta Platforms, Inc.
4
Ohio State University Libraries. Fair Use and Artificial Intelligence 2026 Update
5
Norton Rose Fulbright. AI in Litigation Series: An Update on AI Copyright Cases in 2026
6
U.S. District Court for the District of Delaware. Thomson Reuters v. Ross Intelligence Inc.
7
Skadden. Court Reverses Itself in AI Training Data Case
8
Reed Smith. Court AI Fair Use: Thomson Reuters v. Ross Intelligence
9
CBC News. Anthropic AI Copyright Settlement
10
Classaction.org. Bartz et al v. Anthropic PBC Settlement Notice
11
Writer Beware. Anthropic Copyright Settlement April Update
12
Lieff Cabraser. Authors Secure $1.5 Billion Settlement in Landmark AI Piracy Case
13
AI Lawsuit Tracker. New York Times v. OpenAI
14
The Guardian. US Authors’ Copyright Lawsuits Against OpenAI and Microsoft Combined in New York
15
Bloomberg Law. OpenAI Sued by New Set of Authors Over Training Data Copyrights
16
Baker McKenzie. Case Tracker: Artificial Intelligence Copyrights and Class Actions
17
MeShip Law. Andersen v. Stability AI Litigation Tracker
18
Getty Images Newsroom. Getty Images Issues Statement on Ruling in Stability AI UK Litigation
19
CourtListener. Getty Images (US), Inc. v. Stability AI, Ltd.
20
Georgetown Law Tech Institute. Disney, NBC Universal, and DreamWorks File Major IP Lawsuit Against AI Image Generator Midjourney
21
Music Business Worldwide. Musicians Union Sues UMG and Warner Music
22
Universal Music Group. Universal Music Group and Udio Announce Strategic Agreements
23
Reuters. US Music Publishers Suing Anthropic Make Their Case Against AI Fair Use
24
Baker McKenzie. Concord Music Group, Inc. v. Anthropic PBC
25
IPWatchdog. Amici Back AI Company’s Third Circuit Appeal of Summary Judgment in Thomson Reuters
26
CourtListener. Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc.
27
Baker McKenzie. The Copilot Litigation
28
Digiday. A Timeline of the Major Deals Between Publishers and AI Tech Companies in 2025
29
NPR. New Licensing Deal Highlights the Growing Trend of Media Giants Embracing AI
30
IPWatchdog. CLEAR Act to Establish Notice Requirements for Copyrighted Works in AI Training Data
31
European Parliament. EU AI Act and Copyrights