Anthropic Lawsuits

Plaintiffs. Authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, on behalf of a class of book authors whose works were ingested.

Theory of liability. The complaint alleges Anthropic copied millions of copyrighted books from sources including pirate repositories like LibGen ("Library Genesis") to train Claude, infringing the authors' exclusive reproduction right under the Copyright Act. Anthropic raised fair use as its primary defense.

The Alsup summary-judgment ruling (June 23, 2025). Judge William Alsup split the question into two: training and acquisition. He held that training a generative model on lawfully-acquired books is fair use — the use is transformative, the model does not output the underlying works, and the market harm to authors is speculative. He held that acquisition of pirated copies is not fair use — downloading from a pirate library to build a permanent in-house corpus is straightforward infringement, regardless of what the corpus is later used for. The ruling set the case up for a piracy-damages trial covering more than seven million books Anthropic had acquired through pirated sources.

Settlement (August 2025). Anthropic agreed to pay $1.5 billion — the largest copyright settlement in U.S. history — structured as four installments at roughly $3,000 per eligible work, paid into the settlement fund on October 2, 2025; April 30, 2026; September 25, 2026; and September 25, 2027. Approximately 482,460 books were identified as in-scope. The claim deadline was March 30, 2026, and at the deadline 440,490 of 482,460 eligible works (91.3%) had been claimed by roughly 120,000 authors and rightsholders. The settlement-administration vehicle at anthropiccopyrightsettlement.com publishes the claim mechanics, key dates, and FAQ.

Final approval hearing (May 14, 2026). Judge Araceli Martínez-Olguín moved the fairness hearing to May 14, 2026 at 2:00 p.m. PT (San Francisco federal courthouse, Courtroom 12). At the hearing the court will consider claim rates, opt-outs, and objections from class members; objectors who timely filed may join via Zoom. Unsealed objections cover topics including the exclusion of foreign / non‑US‑registered works, the publisher-versus-author allocation, and the adequacy of class notice. After the hearing the settlement administrator will calculate distributions by June 11, 2026, with payment disbursement expected to begin June 2026 or later (subject to any appeals).

Why it matters. The Alsup fair-use ruling on training is the most-cited piece of LLM-copyright law to date. It established a federal-trial-court line that training on properly-acquired works is fair use while sourcing through piracy is not — a distinction every other AI-copyright defendant now reads for guidance, and one the music publishers re-used as the theory of their January 2026 piracy complaint (see Concord II).

What to watch next. The May 14 fairness hearing and any appeals from objectors. The remaining three settlement installments (April 30, 2026 / September 25, 2026 / September 25, 2027) and the distribution calculations due June 11, 2026. Six authors who opted out of the class have filed individual lawsuits seeking $150,000 per title under the Copyright Act — against Anthropic, OpenAI, Google, Meta, xAI, and Perplexity AI — a parallel track outside the settlement that will play out separately. The Alsup ruling itself continues to be cited in active dockets including Concord Music Group, the Concord-publisher piracy complaint, NYT v. OpenAI, and the Authors Guild v. OpenAI consolidated litigation.

Plaintiffs. Concord Music Group, Universal Music Publishing Group, and ABKCO Music. Filed October 18, 2023 in the Middle District of Tennessee; transferred to the Northern District of California (5:24-cv-03811) where the broader AI-copyright bar is concentrated. Before Judge Eumi K. Lee.

Theory of liability. Two distinct flavors of copyright claim. First, that Anthropic ingested copyrighted song lyrics during training without licensing them (the same training-corpus theory as Bartz, but for music compositions rather than books). Second, that Claude as deployed reproduces those lyrics in response to user prompts — an output-side claim, harder to make on the books side because Claude does not typically reproduce books verbatim, but reachable for short-form lyrics. Plaintiffs sought $75 million plus an injunction.

Preliminary injunction denied (March 25, 2025). Judge Lee denied the publishers' motion for a preliminary injunction. The court held that allegations unidentified users "might" prompt Claude to produce copyrighted lyrics were insufficient to establish the third-party direct infringement that contributory and vicarious theories require, and that the requested injunction was overbroad and not necessary to prevent irreparable harm.

Motion to dismiss granted (March 26, 2025). One day later the court granted Anthropic's motion to dismiss the publishers' claims for contributory and vicarious infringement and intentional removal of copyright management information, with leave to amend. The court found plaintiffs had not pled facts showing actual third-party infringement (i.e., that real users had prompted Claude to generate the copyrighted lyrics) or that Anthropic had the requisite knowledge of any such infringement. The direct-infringement claim on the training side survived.

Procedural posture (2026). Discovery is ongoing. As of January 2026, the court denied Anthropic's motion to clarify a Protective Order amid disputes over alleged information misuse across this case and Bartz (where the same plaintiffs' counsel appeared); Anthropic has separately sought relief from a magistrate-judge order that denied access to pre-suit prompts but required production of post-suit prompts. The Concord piracy follow-on filed January 28, 2026 is now proceeding alongside this case.

Why it matters. The lyrics theory reaches the same fair-use question the Bartz ruling resolved on the training side, but the output-reproduction claim is novel: it tests whether a model that can emit verbatim copyrighted text on demand is itself an infringing instrument. The 2025 dismissal of the secondary-infringement counts narrowed the viable theory significantly — the case now turns on the training-input direct-infringement track rather than the output-reproduction track that drove the early framing.

What to watch next. Whether the court reads the Alsup distinction (training fair use, piracy not) into the music context. Discovery friction over what user-prompt evidence Anthropic must produce. Whether the publishers re-plead the secondary-infringement claims with named users. Settlement is plausible but not signaled.

Plaintiffs and defendants. Concord Music Group, Universal Music Group, and ABKCO Music — the same plaintiff coalition as the 2023 Concord case. Filed January 28, 2026 in the Northern District of California (5:26-cv-00880). Defendants are Anthropic PBC plus, individually, CEO Dario Amodei and cofounder Benjamin Mann.

Theory of liability. The complaint applies the Bartz piracy template to music. The publishers allege that in June 2021, before Anthropic's first product launch, Mann personally used BitTorrent to download approximately five million pirated books from LibGen ("Library Genesis") and PiLiMi ("Pirate Library Mirror") — books that contained the publishers' copyrighted song lyrics, sheet music, and musical compositions — after discussing with Amodei and CSO Jared Kaplan whether to source through piracy rather than license. The complaint alleges Amodei admitted at the time that legitimate licensing channels existed but were rejected as too slow and too expensive. 714 individual works are identified in Exhibit A (including "Wild Horses," "Sweet Caroline," "Bennie and the Jets," "Eye of the Tiger"); a further 20,517 works are identified in Exhibit B. Counts include direct copyright infringement, contributory and vicarious infringement, and DMCA § 1202 violations for removal / alteration of copyright management information.

Damages. The publishers seek roughly $3 billion in statutory damages plus injunctive relief. The complaint also names specific Claude versions (Claude 4.5 Opus, 4.5 Sonnet, 4.5 Haiku) as the trained-on-pirated-works output surface.

Why it matters. This is the first major LLM piracy-acquisition case filed after the Alsup ruling and the Bartz $1.5 billion settlement. The publishers are explicitly using the line Alsup drew — train freely on lawfully-acquired works, but piracy-sourced acquisition is straightforward infringement — as the theory of the case. Naming Amodei and Mann individually is the more aggressive procedural choice; if the court allows the individual claims to survive, the personal-liability surface for AI-lab founders expands materially. The case sets up a direct test of whether the Bartz piracy theory carries over to music compositions and whether the settlement template will repeat at music-publisher scale.

What to watch next. Anthropic's motion to dismiss, particularly on the individual-defendant counts. Any consolidation or coordination with the original 2023 Concord case before Judge Lee. Whether Anthropic pursues an early settlement on the Bartz template (the publishers explicitly framed the filing as a $3 billion claim, which is the kind of public framing that often presages a negotiated outcome).

Plaintiff. Reddit, Inc. Filed June 4, 2025 in San Francisco Superior Court (California state court). Removed by Anthropic to the Northern District of California in July 2025 (3:25-cv-05643). Remanded back to state court March 30, 2026.

Theory of liability. Not copyright. Reddit alleges Anthropic violated Reddit's user agreement and developer terms by continuing to scrape Reddit content after Reddit's commercial-data licensing program (which OpenAI and Google have signed onto) demanded that scrapers either pay or stop. The complaint pleads breach of contract, unjust enrichment, trespass to chattels, tortious interference, and unfair competition under California's Unfair Competition Law (Cal. Bus. & Prof. Code § 17200).

Removal and remand. Anthropic removed from state court in July 2025, arguing Reddit's claims were preempted by the federal Copyright Act and therefore federal-question jurisdiction lay. Reddit moved to remand. The federal court (Judge Trina L. Thompson, N.D. Cal.) signed a tentative remand order on March 20, 2026 and the final order on March 28, 2026, filed March 30, 2026. The court held that none of Reddit's five claims is preempted by copyright: the user-agreement obligations Reddit invokes are qualitatively different from rights granted by copyright law — in particular, Reddit's terms restrict scraping for commercial use, regulate technical-safeguard bypassing, and impose access conditions that copyright does not. Reddit's allegations that Anthropic "bypassed technical safeguards, violated contractual access restrictions, misrepresented its compliance, and exploited Reddit's platform without authorization" sit outside the copyright preemption zone.

Why the remand ruling matters. Beyond venue, the order is the first significant federal opinion holding that platform-TOS / data-licensing claims are not preempted by copyright when they're built around scraping conduct rather than the underlying content's copyright status. AI labs that hoped to fold platform-licensing exposure into the broader fair-use battle have less ground to stand on as a result. The order is being read across to other platform vs. AI-lab disputes.

Why the case matters. This is the highest-profile of the platform-data-licensing suits to date. The Bartz / Concord theories run through copyright; Reddit v. Anthropic runs through contract. If Reddit prevails on the merits, the result is a parallel track of liability that fair-use defenses do not reach — every AI lab that crawled a major social platform has terms-of-service and rate-limit-evasion exposure under the same theory. Reddit has separately signed paid-licensing deals with several other AI vendors; the litigation is leverage as much as it is a damages claim.

What to watch next. The case is now back in San Francisco Superior Court. Trial was previously set for February 14, 2028 in the federal forum; whether that schedule survives the remand is unsettled. Watch for Anthropic's motion-to-dismiss / demurrer in state court and the substantive merits posture on the contract and UCL counts — California state courts apply § 17200 more aggressively than the federal forum did.