AI Copyright Lawsuits Statistics (2026): 40+ Data Points on Court Rulings, Billion-Dollar Settlements, and Case Volume

AI copyright lawsuits statistics 2026: 40+ sourced stats on fair-use rulings, the $1.5B Anthropic settlement and case counts, from the US Copyright Office and RIAA.

Anthropic agreed to pay $1.5 billion to settle Bartz v. Anthropic in September 2025 - the largest copyright recovery in US history, covering 482,460 works at roughly $3,000 apiece (NPR, September 2025; Authors Guild, 2026). That single settlement reframed the financial stakes of AI training disputes, and it is not an outlier: more than 70 AI copyright lawsuits had been filed by the end of 2025, up from around 30 a year earlier (Copyright Alliance, 2025 Year in Review). Courts split sharply in 2025 - training on lawfully acquired books was ruled fair use in two California cases, while a Delaware court and the UK High Court reached narrower or opposite conclusions on related claims. Meanwhile the US Copyright Office concluded that assembling copyrighted works into a training dataset can be prima facie infringement (US Copyright Office, Part 3 Report, May 2025). This analysis consolidates data from the Copyright Alliance, Norton Rose Fulbright, NPR, the Authors Guild, the US Copyright Office, the RIAA, and 15 other primary sources to map the AI copyright litigation landscape as of July 2026.


TL;DR

  • More than 70 AI copyright lawsuits had been filed by the end of 2025, up from around 30 at the end of 2024 (Copyright Alliance, 2025 Year in Review).
  • Anthropic’s $1.5 billion settlement in Bartz v. Anthropic is the largest copyright recovery in US history (NPR, September 2025).
  • The settlement covers 482,460 works at roughly $3,000 per title, and 91.3 percent had been claimed by the March 2026 deadline (Authors Guild, 2026).
  • Anthropic downloaded over 7 million books from pirate sites for a permanent central library (Publishers Weekly, 2025).
  • Thomson Reuters v. Ross was the first US federal ruling to reject an AI fair-use defense, finding more than 2,200 of nearly 3,000 headnotes infringed (Reed Smith, 2025).
  • In re OpenAI consolidated 12 copyright cases into a single MDL in the Southern District of New York in April 2025 (SDNY / In re OpenAI, 2025).
  • OpenAI was ordered to produce a total of 108 million ChatGPT output-log samples in discovery (Norton Rose Fulbright, 2026).
  • Damages in the major-label case against Suno could top $9 billion after labels moved to expand the track list from 560 to 61,026 (Complete Music Update, 2025).
  • Disney and Universal sued Midjourney in June 2025 over 150-plus works, seeking $150,000 each (NPR / CNN, June 2025).
  • 98 percent of authors surveyed said AI companies should ask permission before training on their work (Australian Society of Authors, 2025).
  • The US Copyright Office received more than 10,000 public comments on AI and copyright (US Copyright Office, 2025).
  • The EU AI Act’s training-data transparency and copyright obligations for general-purpose AI took effect on August 2, 2025 (Herbert Smith Freehills Kramer, 2025).

1. The Litigation Landscape by the Numbers

The volume of AI copyright litigation more than doubled in a single year. The signal is not just the raw count but the consolidation: rather than dozens of scattered suits, the largest bloc of author claims against OpenAI and Microsoft is now a single multidistrict proceeding, which concentrates discovery fights and rulings. The most consequential number here is discovery-driven: OpenAI was ordered to produce 108 million ChatGPT output-log samples, a data-disclosure precedent that shapes how plaintiffs prove infringement (Norton Rose Fulbright, March 2026). For the broader generative-AI backdrop, see our generative AI statistics roundup.

MetricValueSource
AI copyright lawsuits filed, end of 202570+Copyright Alliance, 2025 Year in Review
AI copyright cases, end of 2024~30Copyright Alliance, 2025 Year in Review
Initial wave of suits filed in 2023At least 12Ropes & Gray, 2024
Cases consolidated in In re OpenAI MDL12SDNY / In re OpenAI, 2025
MDL centralization (SDNY, MDL No. 3143)April 3, 2025JPML / SDNY court record, 2025
ChatGPT output-log samples ordered in discovery108 millionNorton Rose Fulbright, 2026
Initial de-identified ChatGPT logs ordered20 millionTechCrunch, July 2026

Note: third-party trackers report higher tallies by counting adjacent claims - one AI litigation tracker lists 188 total AI cases against 67 defendants - but those aggregate figures mix copyright with privacy, DMCA and trademark counts (AI Lawsuit Tracker, 2026).

Source: Copyright Alliance - AI Copyright Lawsuit Developments in 2025


2. The Anthropic $1.5 Billion Settlement Benchmark

The Anthropic settlement set the first hard dollar benchmark for training on pirated material. The mechanics matter more than the headline: the fair-use ruling protected training itself, but Anthropic’s downloading of a pirated library did not qualify, and it was that piracy exposure - not the training - that drove the payout. At roughly $3,000 per work across 482,460 titles, the deal implies a per-book price that dwarfs any prior statutory copyright recovery, and 91.3 percent of eligible books were claimed before the March 2026 deadline (Authors Guild, 2026).

MetricValueSource
Total settlement$1.5 billionNPR, September 2025
Works covered482,460Copyright Alliance, 2025
Approximate payout per work~$3,000NPR, September 2025
Books claimed by March 2026 deadline440,490 (91.3%)Authors Guild, 2026
Books Anthropic downloaded from pirate sites7 million+Publishers Weekly, 2025
Class size (rightsholders)Nearly 500,000Authors Guild, 2025
Attorneys’ fee request~$300 million (20%)Norton Rose Fulbright, 2026
Max statutory damages per willful work$150,000US Copyright Act, Section 504

Context: because statutory damages can reach $150,000 per infringed work and the class was potentially in the millions, pre-settlement exposure was estimated in the hundreds of billions - a key reason Anthropic settled rather than face a damages trial (ArentFox Schiff, 2025).

Source: NPR - Anthropic to pay authors $1.5 billion in settlement


3. The Fair Use Scorecard: How Courts Have Ruled

There is no single answer yet on whether training on copyrighted works is fair use - the 2025 rulings point in different directions depending on how the works were acquired and how the plaintiffs argued market harm. The pivotal nuance came from Judge Chhabria in Kadrey v. Meta: he ruled for Meta but warned the 13 authors lost because they failed to develop a market-dilution record, not because training is inherently lawful (Goodwin, June 2025). That reasoning is why plaintiffs’ lawyers now foreground market harm.

CaseRuling / OutcomeSource
Thomson Reuters v. Ross (Feb 11, 2025)First federal ruling rejecting AI fair use; 2,200+ of ~3,000 headnotes infringedReed Smith / Davis Wright Tremaine, 2025
Bartz v. Anthropic (Jun 23, 2025)Training on lawful copies is transformative fair use; pirated library is notNorton Rose Fulbright, 2026
Kadrey v. Meta (Jun 25, 2025)Fair use for Meta; 13 authors failed on market-dilution theoryGoodwin, 2025
Getty Images v. Stability AI, UK (Nov 4, 2025)219-page judgment; secondary copyright claim rejected, narrow trademark winUK Judiciary, 2025
Thomson Reuters appealThird Circuit accepted interlocutory appeal Jun 17, 2025IPWatchdog, 2025
Thaler v. Perlmutter (Mar 2, 2026)Supreme Court denied cert; AI-only output not copyrightable without human authorshipSupreme Court / Thaler v. Perlmutter, 2026

Outlier: the UK Getty ruling turned largely on jurisdiction - Getty conceded Stable Diffusion’s training did not occur in the UK - so it is a narrow procedural signal, not a broad fair-use precedent (Latham & Watkins, 2025).

Source: Norton Rose Fulbright - An update on AI copyright cases in 2026


4. Music and Hollywood Enter the Fray

The most extreme damages exposure sits outside the book cases. Because each sound recording carries up to $150,000 in statutory damages and labels are expanding the count of allegedly copied tracks, the music suits scale into the billions faster than any author class. Sony and Universal moved to expand the list of recordings Suno allegedly copied from 560 to 61,026 tracks, pushing potential damages past $9 billion (Complete Music Update, 2025). For the commercial side of these platforms, see our AI music generation statistics; for the image-generation defendants, our AI image generation statistics.

MetricValueSource
RIAA suits filed against Suno and UdioJune 2024RIAA, 2024
Works alleged against Suno (original filing)662Resemble AI / trade press, 2026
Works alleged against Udio (original filing)1,670Resemble AI / trade press, 2026
Proposed expansion of Suno track list560 to 61,026Complete Music Update, 2025
Potential Suno damages if expanded$9 billion+Complete Music Update, 2025
Disney and Universal v. Midjourney complaint110 pages, 150+ worksNPR / CNN, June 2025
Warner Bros. Discovery v. Midjourney filedSeptember 4, 2025Hollywood Reporter, 2025
UMG settlement with UdioOctober 2025RIAA / trade press, 2025

Context: not every label chose litigation to the end - Universal settled with Udio (October 2025) and Warner settled with Suno (November 2025) to co-build licensed platforms, while other labels kept litigating.

Source: RIAA - Record companies bring landmark cases against Suno and Udio


5. What Creators and the Public Think

Survey data shows a near-consensus among creators that consent should precede training - a gap between what courts have (partly) permitted and what rightsholders expect. The demand is consistent across author and visual-artist surveys and spans consent, compensation and credit. In the Australian Society of Authors 2025 survey, 98 percent said AI companies should ask permission before using their work to train models (Australian Society of Authors, 2025). Consent is also the dividing line for voice work - our recap of 2026 voice-cloning legal cases tracks how the same principle plays out for performers’ voices.

MetricValueSource
Authors: AI companies should ask permission98%Australian Society of Authors, 2025
Authors: want compensation for past use92%Australian Society of Authors, 2025
Authors: against use without explicit consent96%Creator survey, 2025
Visual artists: support a compensation scheme73%NAVA, 2025
Visual artists: reported lost income from AI substitution74%NAVA, 2025
Americans: AI firms rarely or never ask permission28%YouGov, 2024
US adults: more concerned than excited about AI~50%Pew Research Center, June 2025

Note: the YouGov consent figures are the most recent publicly available on that specific question (YouGov, 2024) and are flagged as such; the Pew figure is from a June 2025 survey.

Source: Australian Society of Authors - 2025 AI survey results


Policy is moving on two tracks - a US Copyright Office that leans toward rightsholders on training, and an EU regime that now mandates training-data disclosure. The US Copyright Office concluded that training a model to produce content that competes with the originals is at best modestly transformative, undercutting the broad fair-use claims AI developers rely on (US Copyright Office, Part 3 Report, May 2025). The volume of engagement was itself notable: more than 10,000 public comments, about half addressing copyrightability.

MetricValueSource
Public comments received on AI and copyright10,000+US Copyright Office, 2025
Share of comments addressing copyrightability~halfUS Copyright Office, 2025
Part 3 report on generative AI training releasedMay 9, 2025US Copyright Office, 2025
EU GPAI Code of Practice publishedJuly 10, 2025EU AI Act / artificialintelligenceact.eu, 2025
EU training-data transparency and copyright rules in forceAugust 2, 2025Herbert Smith Freehills Kramer, 2025
EU AI Office enforcement of GPAI rules beginsAugust 2, 2026Skadden, 2025
Compliance deadline for pre-existing GPAI modelsAugust 2, 2027WilmerHale, 2025

Context: the EU’s copyright obligation requires general-purpose model providers to honor rightsholder opt-outs under the 2019 Text and Data Mining exception (Article 4(3)) and publish a summary of training data (Clifford Chance, 2025).

Source: US Copyright Office - Copyright and Artificial Intelligence


MetricValueSource
AI copyright lawsuits filed, end of 202570+Copyright Alliance, 2025
AI copyright cases, end of 2024~30Copyright Alliance, 2025
Anthropic settlement (largest in US history)$1.5 billionNPR, 2025
Works covered by Anthropic settlement482,460Authors Guild, 2025
Approximate payout per work~$3,000NPR, 2025
Books claimed by March 2026 deadline91.3%Authors Guild, 2026
Books Anthropic downloaded from pirate sites7 million+Publishers Weekly, 2025
Cases consolidated in In re OpenAI MDL12Norton Rose Fulbright, 2026
ChatGPT output-log samples ordered in discovery108 millionNorton Rose Fulbright, 2026
Westlaw headnotes found infringed (Ross)2,200+Reed Smith, 2025
Getty v. Stability AI UK judgment length219 pagesUK Judiciary, 2025
Proposed Suno track-list expansion560 to 61,026Complete Music Update, 2025
Potential Suno damages if expanded$9 billion+Complete Music Update, 2025
Disney/Universal v. Midjourney works alleged150+NPR / CNN, 2025
Max statutory damages per willful work$150,000US Copyright Act, Section 504
Authors: AI companies should ask permission98%Australian Society of Authors, 2025
Visual artists reporting AI-driven lost income74%NAVA, 2025
US Copyright Office public comments received10,000+US Copyright Office, 2025
EU GPAI transparency and copyright rules in forceAug 2, 2025Herbert Smith Freehills Kramer, 2025
Thaler v. Perlmutter - Supreme Court cert deniedMarch 2, 2026Supreme Court, 2026

Methodology and Sources

Data was gathered in July 2026 from primary court records, government reports, industry-association filings, named litigation trackers maintained by law firms, and creator-body surveys; every figure was traced to the organization that originally published it, and secondary trade-press reporting was used only to reach the underlying primary source.

Sources cited:

Data watch: The Copyright Alliance publishes an annual AI litigation year-in-review (next edition expected early 2027); Norton Rose Fulbright and McKool Smith update their AI litigation trackers on a rolling basis; the Australian Society of Authors and NAVA run periodic creator surveys; Pew Research Center and YouGov refresh AI public-opinion polling regularly; and the US Copyright Office is expected to publish the final version of its Part 3 report. The EU AI Office begins enforcing general-purpose AI obligations on August 2, 2026.

Last updated: July 10, 2026. We review and update this page quarterly as new data is published.

Try VoxBooster — 3-day free trial.

Real-time voice cloning, soundboard, and effects — wherever you already talk.

  • No credit card
  • ~30ms latency
  • Discord · Teams · OBS
Try free for 3 days