Bartz v. Anthropic: $1.5 billion — final. On July 20, Judge Martínez-Olguín issued final approval of the largest copyright settlement in US history. Approximately 500,000 works at an implied rate of roughly $3,100 per book. The 92.77% opt-in rate across 447,576 claimed works made it functionally universal. Attorneys' fees were cut from the requested $187.5 million to $101.5 million. The settlement establishes the first hard price point for AI training on copyrighted books — and every future negotiation will reference it.
Five new cases filed in July 2026 alone:
- Textbook authors v. Meta & Zuckerberg (Jul 2) — Class action alleging Meta used textbooks to train Llama and stripped copyright management information. Named plaintiffs include Michael Sullivan and Kenneth S. Saladin.
- Evox Productions v. Stability AI, Runway, DeviantArt, Hugging Face (Jul 2) — Automotive image database alleges 100,000+ images scraped via LAION-5B dataset.
- Hachette, Cengage, Elsevier, and Scott Turow v. Google (Jul 10) — Class action alleging Google reproduced copyrighted works through Google Play Books and pirate sources to train Gemini. The first major publisher action against Google's AI training.
- Sony Music v. Udio (Jul 20) — Second filing covering 30,117 additional sound recordings, alleging stream-ripping from YouTube. Sony is now the last major label still in active litigation against both Suno and Udio.
- Evox Productions v. Midjourney (Jul 30) — Second Evox action alleging 100,000+ images used via LAION datasets plus direct scraping.
GEMA v. Suno — Germany rules against AI music training. On July 31, the Regional Court of Munich found that Suno's AI model itself contained copies of GEMA's copyrighted works, making the text-and-data mining exception inapplicable. The court also found that stream-ripping songs to train AI did not qualify as transformative use. First European court to rule against an AI music generator on the merits.
India rules for OpenAI — but with caveats. On July 24, the High Court of Delhi denied a preliminary injunction in ANI Media v. OpenAI, finding that OpenAI's use of ANI's news content for training and retrieval-augmented generation qualified for fair dealing under Indian law. The court explicitly noted the ruling was preliminary and would not bind the final outcome.
News Corp countersues Brave AI. On July 21, News Corp filed counterclaims alleging Brave scraped, used, and sold News Corp content to AI companies — after Brave had pre-emptively sued News Corp seeking a fair-use declaration.
Musk v. OpenAI concluded. The most closely watched AI governance case ended May 18, 2026. The Musk wildcard is off the board.
NYT v. OpenAI — amended complaint. In June 2026, the Times amended its complaint to allege Microsoft built a dedicated supercomputer specifically to help OpenAI use Times content without authorization — broadening the theory of liability.
Hong Kong rejects TDM exception. On July 23, the Hong Kong government formally dropped its consideration of a text-and-data-mining exception for AI training — a signal that Asia's major financial center sides with rights holders.
EU AI Act reaches full enforcement. August 2, 2026 — full enforcement for new AI models. Pre-existing models must comply by August 2027. The transparency mandates requiring disclosure of training data are now binding.
The original map below has been updated to reflect these developments. The case-by-case tracking continues.
The Case That Will Set the Precedent
The New York Times v. OpenAI & Microsoft (S.D.N.Y., filed Dec. 27, 2023). Judge Sidney H. Stein presiding.
The Times alleges OpenAI and Microsoft copied millions of Times articles to train large language models without permission. Critically, the Times' legal team developed and submitted evidence that ChatGPT could reproduce near-verbatim excerpts of Times articles when prompted, including distinctive phrasing and accurate proprietary details. This evidence is widely seen as undercutting OpenAI's central "transformative use" fair-use defense.
In April 2025, Judge Stein denied most of OpenAI's motions to dismiss, including motions related to works older than three years and motions to dismiss contributory copyright infringement and trademark dilution claims. The decision allowed the case to advance toward summary judgment and likely trial.
In November 2025, OpenAI was ordered to preserve 20 million ChatGPT conversation logs as part of discovery — a ruling OpenAI publicly described as "an invasion of user privacy" and appealed. On January 5, 2026, Judge Stein affirmed the magistrate judge's order compelling OpenAI to produce the full 20-million-log sample. OpenAI complied.
In June 2026, the Times amended its complaint to allege Microsoft had built a dedicated supercomputer infrastructure specifically to help OpenAI use Times content without authorization — broadening the theory of liability to the computing infrastructure itself.
The 20 million anonymized ChatGPT conversation logs now in discovery are widely expected to determine whether the case settles or goes to trial. If plaintiffs' experts find systematic reproduction of copyrighted content, legal analysts expect the results to dictate settlement terms across the consolidated litigation.
The case is now in pre-trial. Bellwether trial selection expected late 2026. Damages reportedly sought: in the billions. Full strategic analysis of the Times's posture: The Times Bet Against the Answer Engine.
Why this case matters more than the others: A ruling against OpenAI would establish that training large language models on copyrighted content without licenses is, on its face, infringement. That would force every AI company in the U.S. to retroactively license its training data — or face damages calculations across the entire training corpus. The financial implications run into the hundreds of billions.
A ruling in favor of OpenAI would establish that training is transformative fair use — and effectively legalize the existing AI ecosystem in the U.S.
There is no middle outcome that does not reshape the industry.
The Other Publisher Cases
New York Daily News, Chicago Tribune, Denver Post, et al. v. OpenAI & Microsoft (S.D.N.Y.). Eight Tribune Publishing newspapers (now Alden Global Capital-owned) filed a similar suit in 2024. Consolidated with the Times case under Judge Stein.
Hachette, Cengage, Elsevier, and Scott Turow v. Google (S.D.N.Y., Jul. 2026). The first major publisher class action against Google over Gemini training. Alleges copies were made from literary works sourced from Google Play Books and scraped from pirate sources and paywalled websites. Opens a second front for publishers against a second major AI platform.
Dow Jones + New York Post v. Perplexity (S.D.N.Y., Oct. 2024). News Corp subsidiaries filed against Perplexity alleging "content kleptocracy" — unauthorized scraping, indexing of paywalled content, and false quote attribution to WSJ and NY Post reporters.
News Corp v. Brave AI (S.D.N.Y., Jul. 2026). News Corp's counterclaims against Brave AI allege scraping, using, and selling copyrighted news content to AI companies. Filed after Brave pre-emptively sued seeking a fair-use declaration.
Center for Investigative Reporting / Mother Jones / Reveal v. OpenAI & Microsoft (S.D.N.Y., 2024). Allegations focused on copyright infringement and trademark dilution.
The Intercept v. OpenAI (S.D.N.Y., 2024). Notable for being among the first publisher suits to focus on Digital Millennium Copyright Act (DMCA) violations alongside copyright infringement.
Raw Story / AlterNet v. OpenAI (S.D.N.Y., 2024). Similar DMCA framing.
The Authors and Creators
Authors Guild v. OpenAI (S.D.N.Y., Sept. 2023). Class-action suit on behalf of authors. Named plaintiffs include John Grisham, Jonathan Franzen, George R.R. Martin, David Baldacci, Jodi Picoult, George Saunders, and dozens more. Ongoing fact discovery; bellwether trial selection expected late 2026.
Bartz v. Anthropic (N.D. Cal., 2024) — SETTLED. The landmark case. Final court approval July 20, 2026. $1.5 billion settlement covering approximately 500,000 works at an implied rate of roughly $3,100 per book. The 92.77% opt-in rate made it functionally universal. This is now the reference price point for AI training on copyrighted books.
Textbook Authors v. Meta & Zuckerberg (S.D.N.Y., Jul. 2026). Class action alleging Meta used textbooks to train Llama and stripped copyright management information, including distribution of copies via BitTorrent protocol. Notable for naming Zuckerberg personally.
Silverman / Kadrey / Golden v. OpenAI / Meta (N.D. Cal., 2023). Sarah Silverman and other authors. Kadrey v. Meta produced a mid-2025 ruling that AI training on books can be transformative fair use.
Tremblay / Awad v. OpenAI (N.D. Cal., 2023). Author class action focused on book content in training data.
The Music Industry
UMG / Concord / ABKCO v. Anthropic (M.D. Tenn., Oct. 2023). Major music publishers alleging Claude reproduces copyrighted lyrics. Statutory damages sought of at least $75 million at $150,000 per infringed work.
Music Publishers v. Anthropic (~$3 billion, S.D.N.Y., Jan. 2026). Separate action alleging Anthropic obtained more than 20,000 musical works from pirate "shadow libraries" for training. Described by plaintiffs as the largest non-class-action copyright case in US history. Anthropic's motion to dismiss or answer was due August 3, 2026.
Sony Music v. Suno (D. Mass., ongoing). Dispositive motions now set for April 9, 2027. The "summer 2026 fair-use ruling" everyone expected is off the calendar.
Sony Music v. Udio (S.D.N.Y., Jul. 2026). Second filing covering 30,117 additional sound recordings. Alleges stream-ripping from YouTube. Discovery extended to August 25, 2026. Sony is the last major label still in active litigation against both Suno and Udio — Warner settled with both in late 2025, UMG settled with Udio in October 2025.
GEMA v. Suno (Regional Court of Munich, decided Jul. 31, 2026). German collecting society won. The court found that Suno's AI model contained copies of GEMA's works, making the EU text-and-data-mining exception inapplicable. The court also concluded that stream-ripping to train AI was not transformative under US fair-use analysis. First European ruling on AI music training on the merits. Appeal expected.
The Image Industry
Getty Images v. Stability AI (U.S. and UK actions, multiple filings 2023–2024). Getty alleges Stability AI used millions of Getty-licensed images to train Stable Diffusion. UK High Court ruling in November 2025 was largely favorable to Stability AI. US case in early stages — next case management conference November 5, 2026.
Evox Productions v. Stability AI, Runway, DeviantArt, Hugging Face (Jul. 2026). Automotive image database alleges 100,000+ images scraped via LAION-5B. Direct and contributory infringement plus CMI violations.
Evox Productions v. Midjourney (Jul. 2026). Second Evox action — same LAION-5B allegations plus direct scraping.
Andersen et al. v. Stability AI / Midjourney / DeviantArt (N.D. Cal., 2023). Class action on behalf of artists.
Studio Ghibli, Disney, Universal v. Midjourney (June 2025). The first major film-studio lawsuit against an AI image generator.
The Privacy Class Actions
P.M. v. OpenAI and consolidated cases (N.D. Cal.). Class actions alleging OpenAI scraped personal information from the internet to train AI models without consent. Claims under the California Consumer Privacy Act (CCPA) and the federal Electronic Communications Privacy Act. Settlement talks reportedly in advanced stages.
The International Front
EU AI Act — full enforcement reached August 2, 2026. New AI models must comply now. Pre-existing GPAI models must comply by August 2, 2027. The transparency mandates requiring disclosure of training data sources are binding. First enforcement actions expected Q4 2026.
ANI Media v. OpenAI (High Court of Delhi, Jul. 2026). India's first major AI training ruling favored OpenAI — the court found training and RAG use of news content qualified for fair dealing under Indian law. But the court explicitly said the ruling was preliminary and would not bind the final outcome. The case continues.
Hong Kong rejects TDM exception (Jul. 2026). Hong Kong's government formally dropped consideration of a text-and-data-mining exception for AI training. Pursuing best-practice guidelines instead — a signal that Asia's major financial center sides with rights holders.
Italian Data Protection Authority v. OpenAI (ongoing). GDPR-based actions over data scraping.
Multiple UK proceedings against Stability AI, OpenAI, and other AI companies over UK-specific copyright provisions.
What's at Stake — The $50 Billion Map
The cumulative legal exposure across active AI copyright and privacy cases now exceeds $50 billion in damages sought. Three potential outcomes shape the next 24 months:
Outcome A — Sweeping wins for AI companies. Courts rule training is transformative fair use. Existing AI training corpora are legalized. Damages claims dismissed or sharply reduced. Publishers retain only prospective licensing leverage.
Outcome B — Sweeping wins for publishers and creators. Courts rule training is infringement. AI companies must retroactively license training data or face damages. The industry faces hundreds of billions in liability.
Outcome C — Mixed settlement landscape. Courts establish narrow rulings on specific facts. Cases settle individually. A new licensing ecosystem emerges incrementally.
The Anthropic settlement — $1.5 billion, final — is Outcome C in action. The reference price is now set. The question is whether everyone else settles at that rate or fights for a different one.
Related: The Times Bet Against the Answer Engine · AI Communications Coverage · Legal at Everything-PR
