US copyright lawsuits against AI companies hit 140 on September 1, 2026, up from 45 at the end of June 2025, based on the case map maintained by law professor Edward Lee. This page covers filing volume, the $1.5 billion Anthropic settlement now moving to payout, what rights holders actually book from data licensing, and the EU fines that became enforceable in August 2026.
AI Training Data Licensing Lawsuit Statistics
How Many AI Training Data Lawsuits Have Been Filed?
The count measures cumulative US copyright lawsuits filed against AI companies since January 2023, not active cases or trials.
Filings roughly tripled in fourteen months, from 45 on June 30, 2025 to 140 on September 1, 2026. The newest entry is a music publisher’s claim against Suno and Bright Data.
| Snapshot date | Cumulative US suits |
|---|---|
| June 30, 2025 | 45 |
| September 20, 2025 | 51 |
| October 24, 2025 | 56 |
| February 7, 2026 | 81 |
| March 9, 2026 | 89 |
| April 9, 2026 | 100 |
| June 1, 2026 | 114 |
| September 1, 2026 | 140 |
Source: Chat GPT Is Eating the World, US map of copyright suits v. AI companies, June 2025 to September 2026
The same tracker counted at least 30 copyright cases outside the United States when it passed 100 US filings on April 9, 2026. Sixteen of the US suits sit in one consolidated docket, In re: OpenAI, Inc. Copyright Infringement Litigation in the Southern District of New York, so the headline count runs ahead of the number of separate trials. For context on how quickly the underlying products scaled, see the broader AI adoption figures.
AI Training Data Lawsuit Settlement Statistics: Bartz v. Anthropic
Judge Araceli Martínez-Olguín granted final approval to the $1.5 billion Bartz v. Anthropic settlement on July 20, 2026 in a 23-page order. It is the largest recorded settlement of a US copyright case.
Class counsel reported that 440,490 of the 482,460 eligible works on the Works List had been claimed as of April 16, 2026. Eleven days before the March 30 deadline the figure had been 264,809.
| Settlement item | Amount or figure |
|---|---|
| Gross settlement fund | $1,500,000,000 |
| Attorneys’ fees requested | $187,500,000 (12.5% of fund) |
| Attorneys’ fees awarded | $101,561,111 |
| Litigation expenses requested | $2,779,950.26 |
| Cost reserve requested | $18,220,000 |
| Service award per named plaintiff | $15,000 (reduced from $50,000 requested) |
| Objections filed and overruled | 53 |
| Opt-outs | 350, covering 1,802 works |
Source: Authors Guild summary of the March 19, 2026 motion, and the court’s final approval order of July 20, 2026
The court put the estimated per-work payment at roughly $3,000 and noted that this is four times the statutory minimum for willful infringement. Anthropic downloaded more than 7 million books from pirate libraries, so the certified class covers a fraction of the corpus at issue.
When Do Authors Get Paid?
Anthropic funds the settlement in four instalments rather than one transfer. A status report filed September 2, 2026 set the first payment at approximately $2,203.56 per work, payable on or before November 15, 2026, drawn from $1,083,000,000 held in escrow including expected interest.
Source: Authors Guild, April 17, 2026, on the Anthropic payment schedule; Bartz class counsel status report, September 2, 2026
A second distribution follows the remaining scheduled payment of $450 million plus interest. Class counsel has not published a per-work figure for that tranche.
AI Training Data Licensing Revenue Statistics at Rights Holders
Three public companies break out a line tied to data or AI licensing. Each defines that line differently, so the figures sit side by side rather than adding up.
| Company and reported line | Period | Revenue | Change |
|---|---|---|---|
| Reddit, Other revenue | FY2025, ended Dec 31 2025 | $140 million | +22% |
| Shutterstock, Data, Distribution and Services | FY2025, ended Dec 31 2025 | $203.3 million | +16% |
| Wiley, AI revenue | FY2026, ended Apr 30 2026 | $49 million | +23% |
| Taylor & Francis, non-recurring data access | FY2024, ended Dec 31 2024 | $75 million-plus | Lower in 2025 |
Sources: Reddit Q4 and full year 2025 results, February 5, 2026; Shutterstock full year 2025 results, February 2026; Wiley fourth quarter and fiscal 2026 results, June 16, 2026; Informa 2025 full-year results, March 2026
Reddit’s Other revenue reached $140 million against $2.2 billion of total 2025 revenue, and the line grew slower than advertising, which rose 74% to $2.1 billion. Wiley said lifetime AI revenue passed $110 million by the end of fiscal 2026. Shutterstock’s segment is the lumpiest of the three because it recognises revenue when metadata is delivered.
Source: Shutterstock quarterly reports and full year 2025 results, Data, Distribution and Services revenue by quarter
Grand View Research values the global AI training dataset market at $3.2 billion in 2025 and $3.9 billion in 2026, projecting $16.3 billion by 2033 at a 22.6% CAGR. That figure measures vendor revenue across dataset and labelling suppliers under the firm’s own scope definition, which is a different thing from publisher licensing income. Spending on AI silicon and hardware dwarfs both, and regional venture funding data shows where the capital lands.
AI Training Data Lawsuit Rulings: What Courts Have Decided
Four decisions carry most of the precedential weight so far. None of them settles whether training on copyrighted works infringes as a general matter.
| Decision | Date | Court | Holding |
|---|---|---|---|
| Bartz v. Anthropic | June 23, 2025 | N.D. Cal. | Training on lawfully acquired books is fair use; storing pirated copies is not |
| Kadrey v. Meta | June 2025 | N.D. Cal. | Summary judgment for Meta on the named authors’ training claims |
| Getty Images v. Stability AI | November 4, 2025 | England and Wales High Court | Model weights are not an infringing copy; limited trade mark finding on watermarks |
| GEMA v. Suno | July 31, 2026 | Munich Regional Court I | Training and outputs infringe as to six works; judgment not final |
Sources: court judgments and law firm analyses, June 2025 to July 2026
Discovery is its own cost line. On January 5, 2026 Judge Sidney Stein affirmed an order requiring OpenAI to produce 20 million de-identified ChatGPT conversation logs; plaintiffs had originally sought 120 million. On September 2, 2026 the Justice Department filed a statement of interest in that litigation arguing that training is fair use. Reading those numbers alongside ChatGPT’s usage and revenue data gives a sense of the scale involved.
AI Training Data Licensing Rules and Penalties in the EU
Europe regulates disclosure rather than damages. Providers of general-purpose AI models must publish a summary of training content on a Commission template, and the AI Office gained the power to fine them in August 2026.
| Measure | Date or figure |
|---|---|
| Training-content summary template adopted | July 24, 2025 |
| GPAI obligations applicable | August 2, 2025 |
| Commission enforcement powers applicable | August 2, 2026 |
| Compliance deadline for pre-existing models | August 2, 2027 |
| Maximum fine, Article 101 | 3% of annual total worldwide turnover or €15,000,000, whichever is higher |
Source: European Commission, EU AI Act Article 53(1)(d) template and Article 101, July 2025 to August 2026
The percentage prong is the operative cap for large providers, since 3% of worldwide turnover exceeds €15 million once revenue passes roughly €500 million. Adoption of the models under scrutiny keeps climbing, per enterprise generative AI deployment data and figures on the conversational AI market.
Music rights holders took the licensing route instead. Universal Music settled with Udio in late October 2025 with a compensatory payment plus a licence for a new platform. Warner Music settled with Udio in mid-November 2025, then became the first major label to settle with Suno, which acquired Warner’s Songkick. None of the three disclosed terms, and Sony Music has not settled. For scale on the providers themselves, see figures on Meta’s assistant reach.
FAQs
How many AI training data lawsuits have been filed?
140 cumulative US copyright lawsuits against AI companies as of September 1, 2026, plus at least 30 outside the United States as of April 2026, per the Chat GPT Is Eating the World case map.
What is the largest AI training data settlement?
Bartz v. Anthropic at $1.5 billion, approved July 20, 2026. Attorneys’ fees were cut to $101,561,111 from the $187,500,000 requested. It is the largest recorded US copyright settlement.
How much does Reddit earn from data licensing?
Reddit reported $140 million in Other revenue for FY2025, up 22% year over year. Reddit does not break out data licensing separately; the line also carries other non-advertising income.
Has Reddit sued an AI company over training data?
Yes. Reddit v. SerpApi and Perplexity AI was added to the US case map in October 2025, when the tracker’s cumulative total stood at 56 filings.
What penalties apply to undisclosed training data in the EU?
Under Article 101 of the AI Act, up to 3% of annual total worldwide turnover or €15,000,000, whichever is higher. Commission enforcement powers became applicable on August 2, 2026.
Sources
https://chatgptiseatingtheworld.com/2026/09/01/latest-map-of-us-copyright-suits-v-ai-cos-total-140/
https://authorsguild.org/news/anthropic-settlement-update-91-percent-of-books-claimed/
https://investor.shutterstock.com/news-releases/news-release-details/shutterstock-reports-full-year-2025-and-fourth-quarter-financial
https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-101
