Data & Research Last verified: 2026-05-28 30+ primary sources · 28 cases, rulings & deals

AI Copyright Lawsuits Tracker (2026)

Every major AI copyright case, settlement, and regulatory ruling affecting authors, publishers, and book-cover designers — with the docket number, current status, dollar amount, next milestone, and a link to the primary source for each. Updated monthly.

Jump to the cases Methodology
$1.5B

Bartz v. Anthropic settlement (Aug 2025) — largest copyright settlement in U.S. history. ~500,000 works, ~$3,000 per work; 91.3% claim rate. Fairness hearing held May 14, 2026; final-approval ruling under advisement.

Source: Authors Guild settlement summary ↗
Sep 8, 2026

Andersen v. Stability AI jury trial start date — first AI copyright jury trial in the U.S. (artist class action, N.D. Cal., Judge Orrick)

Source: CourtListener docket ↗
$3B

Concord/UMG/ABKCO v. Anthropic (Lyrics II, Jan 2026) — music publishers' damages demand for shadow-library training, citing 20,000+ lyrics, compositions, and sheet music files

Source: TechCrunch / N.D. Cal. complaint ↗
20M

ChatGPT logs OpenAI ordered to produce in NYT v. OpenAI — most consequential AI discovery ruling to date (Magistrate Wang, May 13, 2025; affirmed Jan 5, 2026)

Source: Bloomberg Law / SDNY ↗
Late 2026 / 2027

Disney + Universal + Warner Bros. v. Midjourney consolidated trial — first major studios v. AI-image-generator test (C.D. Cal., Judge Kronstadt). Hundreds of characters across Star Wars, Marvel, Pixar, DC, Hanna-Barbera; jury demanded; multi-billion-dollar exposure.

Source: CourtListener — Disney v. Midjourney ↗
€15M / 3%

EU AI Act Article 53 fine ceiling — €15M or 3% of global annual turnover for general-purpose AI providers that fail to publish a training-data summary (in force Aug 2025; full enforcement Aug 2026)

Source: EU Regulation 2024/1689 ↗

About this tracker

Generative-AI copyright litigation is now the biggest single issue facing the publishing industry. This tracker focuses on the cases, rulings, regulations, and licensing deals that materially affect authors, publishers, and book-cover designers — the first $1.5B settlement is paying out, the first AI-copyright jury trial is on the calendar for September 2026, and the EU AI Act's training-data disclosure mandate enters full enforcement in August 2026. For comprehensive coverage of news-publisher, music, and non-Western AI cases, see BakerHostetler's tracker, CourtListener, or Copyright Alliance's year-in-review.

Each entry below names the parties, court, case number, judge, current status, dollar amounts where pled, the next milestone, and a primary-source link (CourtListener docket, court PDF, judiciary website, official press release, or Authors Guild summary). Adjacent items — settlements, regulatory mandates, model contract clauses — are included where they shape the precedent landscape.

If you cite something from this page, please cite the underlying docket or judgment, not us. We're a tracker; the courts are the source.

1. Author class actions vs. LLM providers (U.S.)

Book-author copyright suits against the major LLM developers are concentrated in two venues: the Northern District of California (where the Bartz settlement and the Kadrey v. Meta fair-use ruling came down) and the Southern District of New York (where the MDL 3143 consolidation now bundles the Authors Guild, Tremblay, Silverman, Chabon, Alter/Sancton, and Basbanes cases). Bartz is the global reference point for the legal-acquisition / pirated-acquisition split, but the first AI-training fair-use ruling came earlier — in Thomson Reuters v. Ross Intelligence (D. Del., Feb 2025), where Judge Bibas rejected fair use in a B2B legal-search context. Every subsequent case briefs around both Bibas and Alsup.

Foundational fair-use ruling — first AI-training summary judgment

Thomson Reuters v. Ross Intelligence

May 6, 2020 · D. Del. · 1:20-cv-00613 · Judge Stephanos Bibas

Fair use rejected on summary judgment

First U.S. ruling rejecting fair use for AI training data (B2B legal-search context)

Status (May 2026)
Plaintiff's partial summary judgment granted Feb 11, 2025. Ross Intelligence ceased operations 2021; judgment as to liability now stands. Damages and any appeal proceedings pending.
Key rulings & events
  • Feb 11, 2025 Judge Bibas partial summary judgment for Thomson Reuters: Ross's use of Westlaw headnotes to train its AI legal-research tool was not fair use. Court found the use was non-transformative (Ross built a competing legal-research product) and caused market harm. The earlier 2023 ruling that had favored Ross on fair use was vacated.
  • 2021 Ross Intelligence wound down operations citing the cost of litigation; case proceeded to judgment without commercial defendant.
Next milestone
Damages proceedings and any appeal to the Federal Circuit.
Why it matters
The first U.S. ruling on AI-training fair use — and the first to reject the defense. Cited in every subsequent brief, including the Bartz opposition and the U.S. Copyright Office Part 3 Report. Ross's facts (a B2B competitor product, copying for competitive purpose) differ from generative-AI defendants, but the Bibas framework — non-transformative use + market harm — is now the template every plaintiff invokes.

"It is undisputed that Ross's use was commercial. Even though Ross built a research tool, not a generative AI tool, that is enough."

— Judge Stephanos Bibas (Feb 11, 2025 summary judgment)

Wired — first major fair-use ruling CourtListener docket
Class action — settled

Bartz v. Anthropic PBC

Aug 19, 2024 · N.D. Cal. · 3:24-cv-05417 · Judge William H. Alsup

$1.5B settlement

~500,000 works · ~$3,000 per work · 91.3% claim rate

Status (May 2026)
Fairness hearing held May 14, 2026; final-approval ruling under advisement as of May 28, 2026. Claim deadline expired March 30, 2026 with 440,490 of 482,460 eligible works claimed (91.3%).
Key rulings & events
  • Jun 23, 2025 Judge Alsup partial summary judgment: training on legally acquired books = fair use ("spectacularly transformative"); retaining 7M+ pirated LibGen/PiLiMi copies in a "central library" = infringement and proceeds to trial.
  • Aug 25, 2025 Parties announced $1.5B class settlement — largest U.S. copyright settlement on record.
  • Sep 25, 2025 Preliminary approval granted.
  • Mar 30, 2026 Claim filing deadline expired with a 91.3% claim rate.
Next milestone
Final approval order expected after May 14, 2026 fairness hearing. First author payments anticipated June 2026 or later (pending appeals). Payment schedule: $300M post-approval, $450M by Sep 25 2026, $450M by Sep 25 2027.
Why it matters
First U.S. settlement on AI-training copyright. The Alsup ruling's legal-acquisition / pirated-acquisition split is cited in the Concord Lyrics II complaint, the GEMA Munich ruling, and the Carreyrou opt-out filings.

"The technology at issue was among the most transformative many of us will see in our lifetimes."

— Judge William Alsup (Jun 23, 2025 summary judgment)

Authors Guild settlement summary Settlement portal (claim & payment dates) CourtListener docket
Multidistrict litigation — active discovery

In re OpenAI, Inc. Copyright Infringement Litigation (MDL 3143)

MDL transfer Apr 3, 2025 · S.D.N.Y. · 1:25-md-03143 · Judge Sidney H. Stein

Statutory damages up to $150,000 per work

Consolidates Authors Guild + Alter + Sancton + Basbanes + Tremblay + Silverman + Chabon author class actions

Status (May 2026)
Active discovery in the consolidated class action (Baldacci v. OpenAI). Defendants OpenAI, Microsoft. Member cases formerly in N.D. Cal. transferred April 2025; bellwether selection pending.
Key rulings & events
  • Apr 4, 2025 Judge Stein denied OpenAI/Microsoft motion to dismiss; plaintiffs adequately pled prima facie copyright infringement on both training-input and output theories.
  • Jun 13, 2025 Consolidated Class Action Complaint filed (Baldacci caption — David Baldacci, Mary Bly, Michael Connelly, Sylvia Day, Jonathan Franzen, John Grisham, Elin Hilderbrand, George R.R. Martin, Jodi Picoult, Douglas Preston, George Saunders, Scott Turow, plus Pulitzer-winning nonfiction authors).
  • Oct 2025 Judge Stein denied OpenAI MTD on direct, vicarious, and contributory infringement claims based on ChatGPT outputs.
  • Mar 10–13, 2026 Discovery status conference and settlement conference held.
Next milestone
Ongoing fact discovery; bellwether trial selection expected late 2026.
Why it matters
The largest consolidated author-class action against an LLM provider. Any ruling on the consolidated complaint is broadly precedential across the industry.

"Systematic theft on a mass scale."

— Consolidated complaint, characterising OpenAI's training-data practices

Authors Guild v. OpenAI (CourtListener) MDL 3143 master docket S.D.N.Y. MTD opinion (Apr 4, 2025)
Class action — defense summary judgment

Kadrey, Silverman, Golden et al. v. Meta Platforms

Jul 7, 2023 · N.D. Cal. · 3:23-cv-03417 · Judge Vince Chhabria

Fair-use win for Meta — but narrow

13 named author plaintiffs; class certification was never pursued

Status (May 2026)
Post-summary-judgment. Direct training claims dismissed on fair use; distribution-based claims (Meta's alleged seeding/uploading via torrents during LibGen ingestion) survive. Plaintiffs evaluating appeal.
Key rulings & events
  • Nov 2023 MTD ruling: most ancillary claims (DMCA, negligence, unjust enrichment, UCL) dismissed; direct infringement survived.
  • Jun 25, 2025 Judge Chhabria granted Meta summary judgment on fair use: training LLaMA on plaintiffs' books was "highly transformative" but plaintiffs failed to present admissible market-dilution evidence. Distribution claims allowed to proceed.
Next milestone
Discovery on distribution claims; possible appeal of fair-use ruling.
Why it matters
First U.S. ruling granting fair use to a major LLM training corpus drawn from pirated shadow libraries — but the court explicitly limited the holding to these 13 plaintiffs, leaving the market-dilution theory open for future classes.

"In the grand scheme of things, the consequences of this ruling are limited. This ruling does not stand for the proposition that Meta's use of copyrighted materials to train its language models is lawful."

— Judge Vince Chhabria (Jun 25, 2025)

CourtListener docket
Class action — pleading stage

Bird v. Microsoft Corp. (Megatron lawsuit)

Jun 24, 2025 · S.D.N.Y. · MDL 3143 · Judge Sidney H. Stein

Up to $150,000 per work

Targets Microsoft's Megatron-LM separately from joint OpenAI/Microsoft suits

Status (May 2026)
Tagged into MDL 3143; pleading stage as of May 2026.
Key rulings & events
  • Jun 24, 2025 Complaint filed by Kai Bird, Jonathan Alter, Mary Bly, Eugene Linden, Daniel Okrent, Hampton Sides, Jia Tolentino, Victor LaValle, Rachel Vail, Simon Winchester. Cites Microsoft's HarperCollins licensing deal as evidence Microsoft "knew" books required licensing.
Next milestone
MDL discovery proceedings; bellwether selection.
Why it matters
First standalone author action targeting Microsoft's own LLM stack, anchored on Microsoft's existing book-licensing footprint as proof of knowing infringement.
Publishers Marketplace coverage
Multi-defendant author action — pleading stage

Carreyrou et al. v. Anthropic, OpenAI, Google, Meta, xAI, Perplexity

Dec 22, 2025 · N.D. Cal. · 25-cv-10897

Up to ~$900,000 per book

$150,000 statutory damages × 6 defendants

Status (May 2026)
Initial pleadings; defendants' responses pending. Plaintiffs include Bartz settlement opt-outs led by Pulitzer winner John Carreyrou (Bad Blood).
Key rulings & events
  • Dec 22, 2025 Complaint filed alleging defendants copied authors' books from LibGen, Z-Library, and OceanofPDF.
Next milestone
Initial case management conference and defendant motions to dismiss.
Why it matters
First individual book-author copyright suit against xAI and against Perplexity. Signals that high-profile authors who opted out of class settlements (Bartz) are now stacking statutory damages across multiple defendants.

"A deliberate act of theft."

— Complaint

TechCrunch coverage Bloomberg Law

Bartz applies only to Anthropic and only to past acts; it does not license future training and does not settle claims accruing after Aug 25, 2025. Kadrey applies only to 13 named plaintiffs. Every other author class action is still live.

2. News & media v. AI providers

The NYT v. OpenAI discovery framework matters for every AI training-data lawsuit — the court-ordered preservation and production of ChatGPT logs sets the template that the book-author cases (Bartz, MDL 3143, Kadrey) all operate against. The broader news-publisher litigation cluster (Daily News, CIR, Intercept, Raw Story, Dow Jones v. Perplexity, NYT v. Perplexity, Ziff Davis) and the OpenAI publisher licensing dealflow (News Corp $250M, Axel Springer, FT, Le Monde, Condé Nast, Hearst, plus the 2025 cohort) follow related patterns; for case-by-case coverage of those, see BakerHostetler's tracker or CourtListener.

News-publisher case — active discovery (MDL 3143)

The New York Times v. OpenAI / Microsoft

Dec 27, 2023 · S.D.N.Y. · 1:23-cv-11195 · Judge Sidney H. Stein; Magistrate Judge Ona T. Wang

20M ChatGPT logs ordered produced

NYT seeks "billions" in statutory damages

Status (May 2026)
Active discovery within MDL 3143. The court-ordered chat-log production is the most consequential AI discovery ruling to date.
Key rulings & events
  • May 13, 2025 Magistrate Judge Wang ordered OpenAI to preserve and segregate all output log data.
  • Jun 26, 2025 Judge Stein affirmed the preservation order over OpenAI's privacy objections.
  • Sep 26, 2025 Order modified — OpenAI no longer must preserve newly created logs going forward, but must retain everything already saved.
  • Jan 5, 2026 Judge Stein affirmed Wang's order compelling OpenAI to produce the entire 20-million-log anonymized sample, rejecting OpenAI's hand-picked subset.
Next milestone
Continued discovery production through 2026; trial date not yet set.
Why it matters
Establishes that AI providers may be required to retain user chat logs for litigation under Federal Rule of Civil Procedure 37, with downstream privacy and product-design implications. OpenAI publicly objected to the order in June 2025; the framework was cited in subsequent discovery rulings in Dow Jones v. Perplexity and Ziff Davis v. OpenAI.
OpenAI public response to NYT data demands Bloomberg Law — 20M log production affirmed

Britannica + Merriam-Webster filed against Perplexity in Sep 2025 and against OpenAI in Mar 2026 — reference-publisher litigation against AI is now a distinct sub-track. The 2025–2026 wave of OpenAI publisher licensing deals coexists with active OpenAI publisher litigation.

3. Music v. AI generators

Lyrics are text — so rulings on AI training of song lyrics apply directly to book-text quotation and to the question of whether AI outputs of copyrighted text fragments are actionable. The Concord music publishers' case against Anthropic is the lead vehicle. For coverage of pure music-generator litigation (Suno, Udio — including the UMG–Udio settlement that set the first publicly disclosed AI-music royalty rate at $0.002–$0.005 per generation), see specialist trackers.

Lyrics training — cross-MSJ pending + follow-on piracy case

Concord, UMG, ABKCO v. Anthropic (Lyrics)

Oct 18, 2023 (TN); transferred 2024 · N.D. Cal. · 5:24-cv-03811 · Judge Eumi K. Lee

~499 works · $3B follow-on case (Jan 2026)

Up to $75M originally sought; Lyrics II complaint (Jan 29, 2026) cites 20,000+ shadow-library files for damages "in excess of $3 billion"

Status (May 2026)
Cross-motions for partial summary judgment pending after Mar 23, 2026 publishers' motion. Output guardrails in place by stipulation. A separate Lyrics II complaint was filed Jan 29, 2026 by the same publishers — built on Bartz v. Anthropic discovery that surfaced 20,000+ songs, lyrics, compositions, and sheet music files in Anthropic's shadow-library training data.
Key rulings & events
  • Mar 2025 Judge Lee denied publishers' preliminary injunction — plaintiffs did not prove the existing licensing market was undercut. Parties agreed to maintain Anthropic guardrails preventing Claude from reproducing publisher lyrics.
  • Jan 29, 2026 Lyrics II complaint filed in N.D. Cal. — same publishers, damages "in excess of $3 billion," citing Bartz-discovered shadow-library files.
  • Mar 23, 2026 Publishers filed for partial summary judgment in Lyrics I, citing Anthropic's admission that "at least one Claude model was trained on a dataset containing the lyrics to at least one hundred (100) of Publishers' Works."
Next milestone
Lyrics I summary judgment ruling expected mid-2026; Lyrics II in pleadings.
Why it matters
First major lyrics-training case. The fair-use ruling will define the floor for AI music licensing market rates — and the same logic applies to any AI-generated text that reproduces copyrighted song lyrics, including in books.
CourtListener docket Music Business Worldwide — MSJ filing TechCrunch — $3B Lyrics II case

4. Visual artists & image generators (book-cover relevance)

The visual-artist cases matter to publishers and indie authors because they determine the legal status of AI-generated book covers. Two storylines run in parallel: the U.S. Andersen / Getty / Disney–Universal–Warner cluster heading toward trials in late 2026 and 2027, and the U.K. and German rulings (Getty UK, Kneschke v. LAION) that have already given the EU/UK side of the world an early definitive answer.

Visual-artist class action — first AI copyright jury trial

Andersen v. Stability AI

Jan 13, 2023 · N.D. Cal. · 3:23-cv-00201-WHO · Judge William H. Orrick III

Trial Sept 8, 2026

First AI copyright jury trial in the U.S.

Status (May 2026)
Active discovery. Third Amended Complaint filed Feb 27, 2026; answers Mar 13, 2026. Class certification briefing in spring/summer 2026.
Key rulings & events
  • Oct 30, 2023 Judge Orrick granted most of defendants' MTD but let Andersen's direct copyright infringement claim proceed.
  • Aug 12, 2024 Order on First Amended Complaint accepted the "model theory" — that Stable Diffusion itself may be an infringing copy because it embodies compressed transformations of plaintiffs' works. Direct, induced, contributory infringement, and Lanham Act / trademark claims against Midjourney all survived. DMCA §1202(b) CMI claims dismissed with prejudice.
  • Feb 27, 2026 Third Amended Complaint filed.
Next milestone
Jury trial set to begin September 8, 2026.
Why it matters
Bellwether for the entire visual-AI space. If Judge Orrick's "model-as-copy" theory survives trial, every downstream image — including AI-generated book covers produced via Stable Diffusion or its forks — could be deemed a copy derived from an infringing training set.

"Built to a significant extent on copyrighted works… created to facilitate that infringement by design."

— Judge Orrick characterising plaintiffs' allegations (Aug 12, 2024 order)

CourtListener docket Copyright Alliance case overview
Stock-image plaintiff — partial MTD survived

Getty Images v. Stability AI (U.S. refiled)

Aug 14, 2025 · N.D. Cal. · 3:25-cv-06891-TLT · Judge Trina L. Thompson

Up to ~$1.08B exposure

7,216 copyrighted images × statutory damages

Status (May 2026)
Active in early discovery. Original Delaware action terminated Aug 18, 2025 and refiled in N.D. Cal.
Key rulings & events
  • Aug 14, 2025 Delaware case voluntarily dismissed; refiled in N.D. Cal. on same day.
  • Apr 23, 2026 MTD ruling: Getty's DMCA §1202(a) false-CMI claim dismissed without prejudice; direct copyright infringement, trademark infringement (based on Stable Diffusion outputs reproducing the Getty watermark), and California UCL claim all survived.
Next milestone
First Amended Complaint deadline May 7, 2026; Joint Case Management Statement Oct 29, 2026; Case Management Conference Nov 5, 2026.
Why it matters
The watermark-output trademark theory is the closest analog to a book-cover dispute — if a Getty watermark surviving training-to-output is actionable trademark infringement, a cover designer generating an image with a recognisable logo, mark, or signature artifact faces direct liability for the output, separate from any training-data claim.
CourtListener docket Loeb & Loeb — N.D. Cal. MTD analysis
U.K. judgment — model weights are not a "copy"

Getty Images v. Stability AI (U.K. — High Court of Justice)

Jan 16, 2023 · Chancery Division · IL-2023-000007 · [2025] EWHC 2863 (Ch) · Mrs Justice Joanna Smith DBE

Stability AI wins on substance

Up to $1.7B originally sought; only narrow watermark TM claims succeeded

Status (May 2026)
Judgment handed down Nov 4, 2025. Substantively decided in Stability AI's favour.
Key rulings & events
  • Nov 4, 2025 Primary copyright infringement abandoned mid-trial (Getty could not prove training acts occurred in the U.K.). Secondary copyright infringement REJECTED — Mrs Justice Smith held AI model weights are not a "copy" within the meaning of the Copyright, Designs and Patents Act 1988. Trade mark infringement found in extremely narrow scope: certain early Stable Diffusion versions generated outputs bearing the Getty / iStock watermarks, but infringement was "historic and extremely limited." §10(3) trade mark with reputation claim rejected; additional damages refused.
Next milestone
No appeal publicly reported as of May 2026.
Why it matters
First major Western judgment that AI model weights are NOT themselves a copy of training data. For U.K. publishers commissioning AI-assisted covers, training-data infringement claims are very hard to bring in U.K. courts when training occurred abroad — but output-level trade-mark and watermark claims remain viable.

"The Model itself does not store any of those Copyright Works; the model weights are not themselves an infringing copy."

— Mrs Justice Joanna Smith DBE, [2025] EWHC 2863 (Ch)

Judgment — Judiciary.uk
Studios v. AI image generator — pretrial

Disney + Universal + Warner Bros. v. Midjourney (consolidated)

Disney/Universal Jun 11, 2025; WB Sep 4, 2025 · C.D. Cal. · 2:25-cv-05275-JAK-AJR (lead) · Judge John A. Kronstadt

Multi-billion-dollar exposure

Hundreds of characters across Star Wars, Marvel, Pixar, DreamWorks, DC, Hanna-Barbera

Status (May 2026)
Active discovery in 2026. Midjourney elected to answer rather than move to dismiss. Disney and Warner Bros. cases consolidated in early 2026 for all purposes including trial.
Key rulings & events
  • Aug 6, 2025 Midjourney filed Answer asserting affirmative defenses including fair use ("quintessentially transformative"), license, de minimis, DMCA safe harbor. Jury trial demanded.
  • Nov 18, 2025 Pretrial schedule set.
  • May 22, 2026 Informal discovery conference held.
Next milestone
Trial projected late 2026 / 2027.
Why it matters
First major test of output-level copyright liability for AI image generators with respect to recognisable characters. A studios win would establish per-image infringement for AI-generated likenesses of trademarked characters and reject Midjourney's fair-use defense at the merits stage. Book-cover designers commissioning AI work for any property whose IP holder appears on either complaint should track this case directly.

"A bottomless pit of plagiarism."

— Disney/Universal complaint, characterising Midjourney

Disney v. Midjourney (CourtListener) Warner Bros. v. Midjourney (CourtListener)
Visual-artist class action — past MTD

Zhang v. Google (Imagen)

Apr 26, 2024 · N.D. Cal. · 5:24-cv-02531 · Judge Araceli Martínez-Olguín

Six Google AI products in scope

PaLM, GLaM, LaMDA, Bard, Gemini, Imagen

Status (May 2026)
In discovery alongside other Saveri / Butterick-led visual-artist matters.
Key rulings & events
  • Sep 11, 2025 Court ruled plaintiffs plausibly alleged copyright infringement as to six Google AI products. Imagen image-generator claim survived.
Next milestone
Class certification briefing not yet publicly scheduled.
Why it matters
Same legal team and partial plaintiff overlap with Andersen — the Andersen theories are now being extended to Google's Imagen, which underlies Google's enterprise image-generation stack used by some design platforms.
CourtListener docket
Germany — TDM exception ruling, BGH appeal pending

Kneschke v. LAION

2023 · LG Hamburg 310 O 227/23; OLG Hamburg 5 U 104/24 (appeal) · ruling Dec 10, 2025

LAION wins on §60d UrhG TDM-research exception

Plain-English opt-out held not "machine-readable"

Status (May 2026)
Higher Regional Court of Hamburg dismissed Kneschke's appeal Dec 10, 2025. Leave to appeal to the Federal Court of Justice (BGH) granted; Kneschke pursuing BGH appeal — the first time Germany's BGH will rule on AI training and copyright.
Key rulings & events
  • Sep 27, 2024 LG Hamburg ruled for LAION on §60d UrhG scientific-research TDM exception.
  • Dec 10, 2025 OLG Hamburg affirmed: LAION qualifies as an "other institution" conducting scientific research; Kneschke's plain-English TDM opt-out was held invalid because it was not "machine-readable" under Art. 4 DSM Directive / §44b UrhG.
Next milestone
BGH appeal filing and scheduling (2026).
Why it matters
Under current EU/German law, training on web-scraped images is presumptively lawful absent a machine-readable opt-out — so a German cover designer or publisher who used Stable Diffusion is, on this ruling, not generating an infringing work merely because LAION ingested the source images. The BGH appeal could entrench or open EU exposure.
Bird & Bird — OLG ruling analysis

No CONCEPT-art-association class action has been filed (despite frequent rumours). DeviantArt is a defendant in Andersen but has not filed any mirror suit. The Authors Guild has not filed a U.S. class action specifically against AI-generated book covers as of May 2026 — but the April 2026 Model Contract Clauses prohibit AI cover art without express author consent.

5. International litigation (non-U.S.)

The EU-side cases that directly affect authors and publishers in book markets. Germany's GEMA v. OpenAI is the first European court ruling on AI training and memorization-output infringement; France's SNE v. Meta is the first trade-association coordinated EU author/publisher suit. For broader international coverage (Korea, Japan, China, U.K. pre-action), see BakerHostetler's tracker or the Copyright Alliance year-in-review.

First European AI training/output infringement ruling

GEMA v. OpenAI (Germany — song lyrics)

Nov 2024 · Landgericht München I · ruling Nov 11, 2025

GEMA wins at trial level

First European court to hold an AI developer liable for memorization + regurgitation of copyrighted text

Status (May 2026)
Decided in GEMA's favour. OpenAI considering appeal to the Higher Regional Court (OLG Munich).
Key rulings & events
  • Nov 11, 2025 Court held that "memorization" of copyrighted lyrics into model parameters AND "regurgitation" in outputs both constitute copyright reproduction under German law. OpenAI's §44b UrhG TDM-exception defense rejected: GEMA had effectively reserved rights. OpenAI ordered to refrain from reproducing nine specific lyrics under penalty of up to €250,000 or imprisonment per violation; damages, license fees, and information disclosure awarded.
Next milestone
Appeal filings expected mid-2026.
Why it matters
First binding European precedent that AI training and output of memorised copyrighted text is infringement, even where the DSM TDM exception is invoked. The U.S. Bartz reasoning (fair use for legal acquisition / infringement for pirated acquisition) does NOT map cleanly onto the EU regime — EU rights-holders have an additional, narrower path to a win.
GEMA case page Bird & Bird — Munich landmark ruling
First trade-association coordinated EU author/publisher suit

SNE, SNAC & SGDL v. Meta (France)

Mar 6, 2025 · 3rd Chamber, Paris Judicial Court

Removal of unauthorised training datasets + damages

Allegation that Meta failed to comply with EU AI Act Art. 53(1)(d) disclosure

Status (May 2026)
Pending; pleadings phase.
Key rulings & events
  • Mar 6, 2025 France's main publishers' trade body (Syndicat National de l'Édition) plus the authors' and composers' unions filed jointly, alleging "massive use of copyrighted works without authorisation" to train Llama. Sued on "parasitism" theory in addition to direct copyright.
Next milestone
First procedural hearings in 2026.
Why it matters
First trade-association coordinated suit by a major EU publishers' body against a frontier-AI lab over book training data. Sets up the GEMA-style theory specifically for books.
International Publishers Association coverage CMS case tracker

6. Government, regulation & policy

Regulatory action is increasingly the lever — courts move at trial pace, but the EU AI Act, the U.S. Copyright Office reports, and the U.K. Lords' position have already changed what AI providers must disclose and what copyright registrations are available for AI-assisted works.

Government report — pre-publication version in force

U.S. Copyright Office Report, Part 3 — Generative AI Training

May 9, 2025 (pre-publication draft)

Fair use cannot be presumed

Market-impact framework explicitly considers "lost licensing markets"

Status (May 2026)
Pre-publication draft still in force as USCO guidance; final version delayed amid the Register of Copyrights dismissal controversy.
Key rulings & events
  • May 9, 2025 Fair use cannot be presumed for AI training; case-by-case. "Making commercial use of vast troves of copyrighted works to produce expressive content that competes with them in existing markets, especially where this is accomplished through illegal access, goes beyond established fair use boundaries."
Next milestone
Final publication pending.
Why it matters
Cited heavily in the Bartz settlement positioning and underpins Authors Guild's licensing-market arguments. The "lost licensing markets" framework means every disclosed publisher AI deal narrows the fair-use defense for the next defendant.

"The use of entire copyrighted works is less clearly justified in the context of AI training than it was for Google Books."

— U.S. Copyright Office, Part 3 Report (pre-publication, May 9, 2025)

USCO Part 3 PDF
Government report — binding USCO registration practice

U.S. Copyright Office Report, Part 2 — Copyrightability

Jan 29, 2025

Prompts alone do not confer authorship

Human modification, arrangement, or selection of AI outputs CAN be protectable

Status (May 2026)
Final; binding USCO registration policy.
Key rulings & events
  • Jan 29, 2025 Pure AI outputs are not copyrightable. Disclosure of AI-generated content required on registration applications. Drew on 10,000+ public comments from 67 countries.
Next milestone
Continued case-by-case USCO review.
Why it matters
Establishes that purely AI-generated works are uncopyrightable in the U.S. — the legal foundation underlying "Human Authored" certification programs, and the practical reason indie authors using AI heavily may lose copyright protection on their books.
USCO Part 2 PDF
Regulation — in force

EU AI Act Article 53(1)(d) — GPAI Training Data Summary

Regulation 2024/1689; effective Aug 2, 2025; full enforcement Aug 2, 2026; pre-existing models Aug 2, 2027

€15M or 3% of global turnover

Max fine for non-compliance with training-data summary mandate

Status (May 2026)
In force for new GPAI models since August 2025. AI Office compliance verification beginning ahead of August 2026 full enforcement.
Key rulings & events
  • Jul 24, 2025 AI Office published the official template requiring disclosure of data modalities, dataset size, identification of major public datasets, narrative descriptions of licensed / scraped / user-generated / synthetic data sources.
Next milestone
Full enforcement Aug 2, 2026; pre-existing GPAI models must comply by Aug 2, 2027.
Why it matters
First binding global transparency mandate giving copyright owners enforceable visibility into what AI models were trained on. Underpins every EU-side rights-reservation theory.
EU AI Act Article 53 (official) EU template for training-content summary
Parliamentary committee report

U.K. House of Lords — AI, Copyright and the Creative Industries

Mar 6, 2026 · HL Paper 267 · Communications and Digital Committee

Rejects commercial TDM exception

Endorses licensing-first model + mandatory transparency

Status (May 2026)
Final; Government response pending.
Key rulings & events
  • Mar 6, 2026 Recommends mandatory transparency for large AI developers on training content; backs licensing-first; rejects commercial TDM exception. Notes creative industries' £124B gross value added in 2023.
Next milestone
U.K. Government formal response.
Why it matters
Most authoritative U.K. parliamentary statement to date opposing the U.K. IPO's TDM-with-opt-out proposal. Shifts the U.K. political landscape against the EU-style opt-out model.
HL Paper 267 (official)
Industry contract template

Authors Guild Model Contract AI Clauses (April 2026)

April 2026 update

75–85% recommended author share of AI licensing revenue

No-Upload + No-AI-Substantive-Edit + Author Disclosure clauses

Status (May 2026)
Recommended for all new author–publisher contracts. Voluntary adoption.
Key rulings & events
  • April 2026 Publishers may not upload manuscripts or author personal info to consumer-facing AI without written permission. AI cannot substantively edit a manuscript (spell/grammar OK). Authors must disclose any AI-generated text in submission, capped at de minimis percentage. AI narration / translation / cover art requires prior written consent.
Next milestone
Continued industry adoption negotiation.
Why it matters
Defines the de facto industry standard authors and agents now demand in all new deals. Combined with the Human Authored certification, gives authors a contractual + market lever beyond the courts.
Authors Guild Model AI Clauses
Certification program

Human Authored Certification (US + UK)

US public launch Mar 2, 2026; UK launch Mar 10, 2026 (London Book Fair)

3,000+ authors / 5,000 titles certified

Within ~3 months of public launch

Status (May 2026)
Active; Authors Guild (US) + Society of Authors (UK).
Key rulings & events
  • Jan 2025 Beta launch (Authors Guild members only).
  • Mar 2026 Public US + UK launch; bulk certification available to publishers.
Next milestone
Continued enrolment + publisher partnerships.
Why it matters
First scalable mechanism authors can use to differentiate against AI-generated competition — directly relevant to indie authors' visibility in retail listings and reader trust.
Authors Guild Human Authored Human Authored UK

7. Licensing deals & industry positions

Every disclosed publisher AI deal does two things: it generates revenue for the publisher (and sometimes for authors), and it narrows the fair-use defense for the next defendant by establishing a "market rate." The U.S. Copyright Office's Part 3 framework is explicit on this. Deals are listed below with dollar amounts where disclosed and opt-in/opt-out terms where known.

Trade publishing license — first Big-5

HarperCollins–Microsoft (Big-5 trade publisher AI deal)

Announced Nov 2024

$5,000 per book · 50/50 split

Opt-in · non-recoupable · 3-year non-exclusive

Status (May 2026)
Active 3-year deal. Limited to selected nonfiction backlist titles.
Key rulings & events
  • Nov 2024 Author consent required; $5,000 fee per book, split 50/50 ($2,500 author / $2,500 publisher); non-recoupable; outputs limited to ≤200 consecutive words or ≤5% of book's text.
Next milestone
Term expires 2027.
Why it matters
First Big-5 AI licensing deal — sets the $2,500-per-title benchmark every author now references in their own AI clause negotiations. Cited in Bird v. Microsoft as evidence Microsoft knew books required licensing.
Authors Guild deal summary
Academic publishing license — no opt-out

Wiley AI licensing (academic)

FY2024 $23M; Q1 FY2025 $21M

~$44M total disclosed

No author opt-out · royalties per "contractual terms"

Status (May 2026)
Active.
Key rulings & events
  • FY2024–FY2025 Backlist titles ≥3 years old; AI deals drove 14% sales increase to $124M in Q1 FY2025.
Why it matters
Established the no-opt-out academic publisher model — contrasted with HarperCollins's opt-in trade model.
Publishers Weekly
Academic publishing license — no opt-out

Taylor & Francis (Informa) – Microsoft

May 2024

$10M initial + recurring

~3,000 academic journals · authors not informed in advance · no direct compensation

Status (May 2026)
Active 2024–2027.
Key rulings & events
  • May 2024 Non-exclusive access to ~3,000 academic journals ("Advanced Learning Content"). Authors not consulted; no opt-out provided; no direct author compensation.
Why it matters
Most controversial academic AI deal of 2024 — triggered scholarly society protests and a National Communication Association formal statement.
The Bookseller coverage
Academic publishing license

Springer Nature – Google

Jul 2024

$23M one-time payment

Previously published academic papers

Status (May 2026)
Active.
Why it matters
Sets a benchmark valuation for academic AI training content — similar magnitude to Wiley's initial deal.
Nature coverage
Academic publishing position

Cambridge University Press — opt-in AI policy

Aug 2024

Opt-in only

The lone major academic publisher rejecting the blanket-deal model

Status (May 2026)
Active opt-in policy; no major deals publicly announced.
Why it matters
Only major academic press following the opt-in model — gives authors a published policy lever when negotiating with other academic presses.
The Bookseller coverage
Industry position — DSM rights reservation

Penguin Random House — no-AI copyright-page clause

Oct 2024

First Big-5 to reserve DSM TDM rights formally

Applies to all new books and reprints

Status (May 2026)
Active across all new titles.
Key rulings & events
  • Oct 2024 Copyright page now reads: "No part of this book may be used or reproduced in any manner for the purpose of training artificial intelligence technologies or systems." Explicit reservation under EU DSM Art. 4(3) TDM opt-out.
Why it matters
First Big-5 publisher to formally reserve TDM rights — gives PRH legal standing under EU law (and informs every German-language case post-GEMA).
Authors Guild summary

Three of the Big-5 trade publishers (Macmillan, Hachette, Simon & Schuster) still have no public AI training policy as of May 2026 — a vacuum agents and authors are exploiting in new contracts. The Bertelsmann–OpenAI "strategic collaboration" (Jan 2025) is frequently miscited as a book-training license; per Bertelsmann's own statements, it does NOT give OpenAI access to Penguin Random House's book library. Bloomsbury embraced AI licensing publicly in 2025 and raised full-year outlook on its first deal.

Legal principles settled so far

Across all the cases above, three questions matter most to authors, publishers, and cover designers: is style copyrightable, is the model itself a copy of the training set, and what is the per-work damages exposure? As of May 2026, here is what courts have actually decided.

Legal principle

Is artistic or writing "style" copyrightable?

No — not as of May 2026

Style imitation is not infringement; recognisable characters and watermarks are

Status (May 2026)
No U.S. or U.K. court has held that artistic style is itself copyrightable. The Andersen 2024-08-12 order rejected DMCA §1202(b) claims precisely because outputs were not identical to plaintiffs' works — implicitly leaving style imitation outside the protected zone.
Why it matters
Prompting "in the style of [artist]" does not, under any 2024–2026 ruling, automatically infringe. The Studio Ghibli / OpenAI "Ghiblification" controversy of March 2025 — and Japanese trade body CODA's October 2025 demand letter to OpenAI — exists in this legal limbo: it is actionable as unfair-competition or publicity-rights in some jurisdictions, not as copyright.
Andersen v. Stability AI 2024-08-12 order analysis
Legal principle

Is the AI model itself a "copy" of the training data?

Split between jurisdictions

Accepted at pleading stage in N.D. Cal. (Andersen); rejected on the merits in England (Getty UK)

Status (May 2026)
Judge Orrick's August 2024 Andersen order accepted (at pleading stage) the "model theory" — that Stable Diffusion itself may be an infringing copy because it embodies compressed transformations of plaintiffs' works. Mrs Justice Smith's UK ruling [2025] EWHC 2863 (Ch) expressly rejected the analogous secondary infringement theory, holding model weights are not a "copy" within the CDPA 1988.
Why it matters
The U.S. theory is alive but untested at trial — Andersen's September 2026 jury trial will produce the first U.S. verdict. The U.K. theory is closed unless the Court of Appeal reverses Getty.
Andersen Aug 2024 order analysis (Loeb) Getty UK judgment
Legal principle

What is the per-image / per-work damages exposure?

Up to $150,000 per registered work (willful)

Default $30,000; floor $750; innocent infringement reduces to $200

Status (May 2026)
Under 17 U.S.C. §504(c), statutory damages range from $750 to $30,000 per registered work, up to $150,000 if the infringement was willful. No court has yet ruled on whether the cap applies per training image, per output image, or per model release.
Why it matters
Cover designers and indie authors using AI tools are downstream good-faith users. If a designer's AI vendor is later held infringing, the designer's output (the book cover) is a potential downstream act — but the innocent-infringement provision reduces statutory damages to as low as $200/work. The real exposure sits with the AI vendor, not the designer.
17 U.S.C. §504(c)

How we built this tracker

Every case entry links to a CourtListener docket, court PDF, Authors Guild summary, judiciary website, or first-party press release. Coverage from law firms (Bird & Bird, Loeb & Loeb, Mayer Brown, Latham & Watkins, Hogan Lovells, DLA Piper, Susman Godfrey) is included where it summarises a primary-source ruling we link to.

We do not paraphrase rulings — we quote them. Dollar amounts are sourced; where unverifiable from a primary source, we say "not publicly disclosed." Where a ruling has been appealed or modified, the modification is noted with its date.

Coverage scope: U.S. federal cases involving book authors, news publishers, music publishers, visual artists, or studios as plaintiffs against major LLM or image-generator providers; plus non-U.S. cases in jurisdictions with measurable publishing markets (U.K., Germany, France, Japan, South Korea, China); plus government reports, EU regulations, and major Big-5 / academic-publisher licensing deals that shape the precedent landscape.

Not in scope: GitHub Copilot (code, not publishing); deepfake/voice cases not tied to audiobooks; international cases without an English-language primary source we can link to.

Frequently Asked Questions

Was my book in any of these training sets?
If your book was in LibGen, Z-Library, Books3, OceanofPDF, or Pirate Library Mirror (PiLiMi), it is likely in one or more AI training sets. Direct searches: The Atlantic's Books3 database lookup (https://www.theatlantic.com/technology/archive/2023/09/books3-database-generative-ai-training-copyright-infringement/675363/), the Anthropic settlement claim portal (https://www.anthropiccopyrightsettlement.com/) for Bartz-eligible works, and the Authors Guild's AI training-data tools and member resources (https://authorsguild.org/advocacy/artificial-intelligence/). The Bartz claim deadline expired March 30, 2026 — if your book was eligible and unclaimed, you remain a class member of record but cannot file under that settlement; the Carreyrou case (Dec 2025, see Section 1) is the model for opt-out plaintiffs who chose to sue separately.
What does the Bartz settlement mean if my book wasn't claimed?
The Bartz claim deadline (March 30, 2026) has expired. If your book was eligible and unclaimed, it remains a class member of record but cannot file for payment under this settlement. Future settlements with Anthropic for post-Aug 25, 2025 acts, or with other defendants (OpenAI, Meta, Microsoft, xAI, Perplexity, Google), would be separate claim processes.
Can I still sue an AI company if my book was scraped?
Yes, in most cases. Bartz releases past Anthropic acts for class members; it does not release future acts or other defendants. The Carreyrou et al. v. Anthropic / OpenAI / Google / Meta / xAI / Perplexity case (Dec 2025) was filed by Bartz opt-outs and shows the per-defendant statutory-damages stacking strategy.
Are AI-generated book covers legal?
Under current rulings, the cover itself (the output) is not automatically infringing — but two narrower risks exist. (1) Output-level trademark or recognisable-character infringement is actionable (Getty UK watermark ruling; Disney/Universal/Warner Bros. v. Midjourney). (2) If your AI vendor is later held infringing on training data, your output is a downstream act, though the innocent-infringement statutory-damages floor is $200/work. Andersen v. Stability AI (jury trial September 8, 2026) will produce the first U.S. verdict on the underlying "model-as-copy" theory.
Can I copyright a book I wrote with AI?
In the U.S., pure AI outputs are not copyrightable per the U.S. Copyright Office Part 2 Report (January 2025). Human modifications, arrangement, or selection of AI outputs CAN be protectable. You must disclose AI-generated content on the registration application. China's Beijing Internet Court (Sept 16, 2025) ruled the opposite — AI-generated images CAN be copyrighted under Chinese law with sufficient "creative effort." Most other jurisdictions follow the U.S. approach.
Does the EU AI Act apply to me as a non-EU author?
Not directly. It applies to providers of general-purpose AI models placed on the EU market. But it gives you visibility into what those models were trained on — the official training-content summary template (published July 24, 2025) requires disclosure of major datasets. Combined with the EU DSM Art. 4(3) TDM opt-out, this is the primary EU rights-holder lever.
How is the U.K. position different from the U.S.?
Two material differences as of May 2026. First, Mrs Justice Smith's Getty UK ruling [2025] EWHC 2863 (Ch) held AI model weights are not a "copy" — closing the model-as-infringement theory in the U.K. unless reversed on appeal. Second, the U.K. House of Lords' March 2026 report rejects the EU-style commercial TDM exception that the U.K. Government had proposed; the Government has not yet responded. Litigation in the U.K. is presently limited to output-level trademark and database-right claims (Mumsnet).
What's the precedent from Bartz that other cases are testing?
Judge Alsup split the fair-use question: training on legally acquired books = fair use ("spectacularly transformative"); retaining pirated LibGen / PiLiMi copies in a "central library" = infringement. Every other U.S. case is now arguing whether its facts fall on the legal-acquisition side or the pirated-acquisition side. The U.S. Copyright Office's Part 3 Report adds a third overlay — "lost licensing markets," meaning every disclosed publisher AI deal makes the fair-use defense narrower for the next defendant.
What should I do if my publisher signed an AI licensing deal without telling me?
Check your contract for an AI-rights clause. Most pre-2024 contracts are silent on AI training, which creates ambiguity that has been litigated both ways (Wiley + Taylor & Francis no-opt-out; HarperCollins opt-in). The Authors Guild's April 2026 Model Contract Clauses are the de facto industry standard authors and agents now demand. The Authors Guild–Created by Humans partnership offers an author-controlled opt-in alternative to publisher blanket deals.
How often is this tracker updated?
Monthly while major cases are active. Each entry's status reflects the date in the "Updated" badge at the top of the page. For dockets that move week-to-week (NYT v. OpenAI discovery, Andersen pretrial, MDL 3143), check the primary source link for the latest filing.

Changelog

2026-05-28

Initial publish. 28 cases, rulings, and deals across 7 sections plus 3 legal-principle entries — focused on the litigation, regulation, and contract activity that materially affects authors, publishers, and book-cover designers. Includes Bartz v. Anthropic ($1.5B), Thomson Reuters v. Ross (foundational fair-use ruling), MDL 3143 (Authors Guild + Alter + Tremblay + Silverman + Chabon + Basbanes consolidated), Kadrey v. Meta SJ, Bird v. Microsoft (Megatron), Carreyrou multi-defendant (Dec 2025), NYT v. OpenAI 20M-log production order, Concord lyrics case (incl. $3B Lyrics II Jan 2026), Andersen (trial Sept 8 2026), Getty US refiled + Getty UK [2025] EWHC 2863 (Ch), Disney + Universal + WB v. Midjourney consolidated, Zhang v. Google Imagen, Kneschke v. LAION, GEMA v. OpenAI Munich (first European training/output ruling), SNE v. Meta France, US Copyright Office Parts 2 + 3, EU AI Act Art. 53 (€15M / 3% turnover fine ceiling), UK Lords HL Paper 267, Authors Guild Model Clauses + Human Authored certification, HarperCollins ($5K/book opt-in) / Wiley / Taylor & Francis / Springer Nature / Cambridge UP / PRH no-AI clause.

Full sources

All primary sources cited above, grouped by type. Court dockets via CourtListener (Free Law Project) where available.

Court dockets (U.S.)

Why ManuscriptReport tracks this

We build book-marketing reports for authors and publishers working in this legal landscape every day — so we maintain this tracker as part of our own primary-source workflow. If you need that workflow done for one of your books, our marketing reports are here.

See a sample report Pricing