Last reviewed
Jun 9, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v3 · 4,975 words
Add missing citations, update stale details, or suggest a clearer explanation.
Concord Music Group, Inc., et al. v. Anthropic PBC is a federal copyright lawsuit brought by a coalition of major music publishers against Anthropic, the developer of the Claude family of large language models. The publishers allege that Anthropic engaged in systematic copyright infringement by ingesting the lyrics of hundreds of copyrighted musical compositions into Claude's training data without a license, and by causing Claude to reproduce those lyrics in response to user prompts. The action was filed in October 2023 in the U.S. District Court for the Middle District of Tennessee, transferred to the U.S. District Court for the Northern District of California in mid-2024, and remains pending before Judge Eumi K. Lee, with summary judgment briefing scheduled to conclude in June 2026.[1][2][3]
The case is one of the most closely watched matters in the broader wave of generative AI copyright litigation. It is the first such suit specifically targeting song lyrics rather than books, news articles, source code, or images, and the publishers' demands for both training-side and output-side relief have drawn the court into several first-of-their-kind procedural rulings on AI guardrails, irreparable harm, secondary liability, and the Digital Millennium Copyright Act ("DMCA"). As of May 2026, dispositive motions are pending on threshold fair use questions, and a parallel action filed in early 2026 ("Concord II") expands the dispute to allegations that Anthropic used the BitTorrent protocol to acquire books containing the publishers' lyrics from shadow library archives.[4][5]
The Concord I complaint was filed by eight music publisher plaintiffs:[1][6]
The Universal Music Corp., Songs of Universal, Universal Music - MGB NA, Universal Music - Z Tunes, Polygram Publishing, and Capitol CMG entities all sit within the Universal Music Publishing Group ("UMPG") family, the publishing arm of Universal Music Group. Concord Music Group is the publishing entity behind catalogs that include Pulse Music Group, Rodgers and Hammerstein, and many compositions co-administered with major songwriters. ABKCO Music & Records is the longstanding publisher and rights administrator that controls early Rolling Stones publishing rights along with works by Sam Cooke and others. Together, the plaintiffs hold or administer copyrights in many of the best-known compositions in the U.S. and international popular music canon.[2][7]
The publishers are represented by Oppenheim + Zebrak LLP, the firm that has also represented major record labels in cases against generative AI music developers Suno and Udio.[4][7]
Anthropic PBC is a Delaware-incorporated public benefit corporation headquartered in San Francisco, California, founded in 2021 by Dario Amodei, Daniela Amodei, and several former OpenAI researchers. Anthropic develops the Claude family of large language models and serves enterprise and consumer customers through its API and its Claude.ai chat product. Anthropic is represented in the matter by Latham & Watkins LLP, with Joseph Wetzel, Allison Stillman, Andrew Gass, and Sarang Damle among the lead attorneys identified in court filings.[1][8]
The publishers allege that Anthropic built Claude in part by scraping song lyrics from licensed lyric aggregator sites and from third-party datasets that themselves contain licensed and unlicensed lyric data. According to the complaint, Anthropic's training corpus includes both web data scraped from sources such as MusixMatch, LyricFind, and Genius, as well as text drawn from publicly released datasets including Common Crawl and The Pile, each of which the complaint alleges contains unlicensed copies of the publishers' lyrics.[9][10]
The complaint identifies 499 musical compositions in an attached exhibit as a representative sample of works alleged to have been infringed; press coverage and several court orders subsequently rounded this number to "approximately 500" or characterized the works as "at least 500 songs."[7][11] The named compositions include, by way of example:[7][12]
A central feature of the complaint is the allegation that Claude does not merely store the lyrics during training but reproduces them on demand in response to user prompts. The publishers presented examples of Claude generating verbatim or near-verbatim portions of copyrighted lyrics when asked for them directly, when asked to write a song on a related topic, and even when asked for original compositions in a particular artist's style.[7][12]
Specific prompt examples included in the complaint are widely cited in press coverage:[7][12]
In addition to direct reproduction claims, the publishers asserted a claim under 17 U.S.C. section 1202 of the DMCA for the removal of copyright management information ("CMI"). The complaint alleges that Anthropic ran extraction tools over the scraped pages to strip out copyright notices, author credits, and other identifying metadata before storing the cleaned text in its training corpus. The publishers focus on Anthropic's selection of the "Newspaper" Python library for content extraction, alleging that Anthropic's cofounders deliberately chose Newspaper over alternatives such as jusText specifically because Newspaper was more effective at stripping CMI from web-scraped pages.[9][13]
The complaint as filed asserted four causes of action:[1][6]
For the copyright infringement counts, the publishers sought statutory damages of up to $150,000 per infringed work, the maximum available for willful infringement under 17 U.S.C. section 504(c). For the CMI claims, the complaint sought damages of up to $25,000 per violation under 17 U.S.C. section 1203(c)(3)(B), along with injunctive relief, attorneys' fees, and destruction of infringing materials. Early press coverage described the statutory-damages exposure on the 499-song exhibit as on the order of $75 million if every work were found willfully infringed.[11][14]
Anthropic has consistently denied liability and asserted several defenses, the most significant of which is fair use under 17 U.S.C. section 107.
In its public statements and court filings, Anthropic has characterized the inclusion of lyric text in its training corpus as a "transformative" use that produces a large language model capable of a wide variety of tasks rather than a competing lyrics product. Anthropic has analogized its training process to the intermediate copying upheld as fair use in Authors Guild v. Google (the Google Books case) and in software interoperability cases such as Sega v. Accolade and Sony v. Connectix. Anthropic has also pointed to its content filters as evidence that Claude is not designed to function as a lyrics-distribution service.[8][15]
Anthropic separately argued that:
Anthropic also moved to dismiss the original complaint for lack of personal jurisdiction in Tennessee and for improper venue, arguing that the company has no operations, employees, or principal offices in Tennessee and that the connection asserted by the publishers (a Nashville music-industry presence and the residence of certain plaintiff employees) did not satisfy due-process requirements.[16][18]
The case was filed on October 18, 2023, in the U.S. District Court for the Middle District of Tennessee in Nashville and was docketed as Concord Music Group, Inc. et al. v. Anthropic PBC, No. 3:23-cv-01092. The case was assigned to Chief Judge Waverly D. Crenshaw, Jr.[1][18]
On November 16, 2023, less than a month after filing, the publishers moved for a preliminary injunction seeking two forms of relief: (i) an order requiring Anthropic to cease using the publishers' lyrics in any future training of Claude, and (ii) an order requiring Anthropic to maintain technical "guardrails" to prevent Claude from reproducing the publishers' lyrics in outputs.[18][19]
Anthropic responded with a motion to dismiss for lack of personal jurisdiction and improper venue, along with an opposition to preliminary injunctive relief. Briefing extended into 2024, with the court declining to rule on the preliminary injunction motion until the threshold venue issue was resolved.[16][18]
On June 24, 2024, Judge Crenshaw issued a memorandum opinion granting Anthropic's motion in part. The court found that the publishers had not made the required prima facie showing that Anthropic was subject to specific personal jurisdiction in Tennessee, that the alleged in-forum conduct (Anthropic's nationwide distribution of Claude reaching Tennessee residents and the location of certain music-industry plaintiffs in Nashville) did not satisfy the "purposeful availment" standard, and that transfer to the Northern District of California, rather than dismissal, served the interests of justice. The court therefore transferred the case to the Northern District of California, where Anthropic maintains its principal place of business.[16][18][20]
The case was redocketed in the Northern District of California as No. 5:24-cv-03811 and assigned to Judge Eumi K. Lee, sitting in San Jose. (Magistrate Judge Susan van Keulen was later assigned for discovery matters.)[3][16]
On August 1, 2024, the publishers renewed their motion for a preliminary injunction. Anthropic filed an opposition on August 22, 2024, arguing both that the publishers had failed to show likelihood of success on the merits and that they had not shown irreparable harm given Claude's existing content filters.[8][16]
On November 25, 2024, Judge Lee held a marathon oral argument on the renewed motion.[21]
Before the court ruled, the parties entered into a partial settlement that narrowed the preliminary injunction dispute. On December 30, 2024, the parties stipulated, and on January 2, 2025, Judge Lee entered an order under which:[22][23][24]
The publishers publicly described the order as a partial victory; Anthropic described it as confirming that its existing safety measures already addressed any cognizable risk of output-side infringement. Anthropic also stated that its training practices remained "consistent with existing copyright law" through fair use.[22][24]
On March 25, 2025, Judge Lee denied the remainder of the publishers' motion for a preliminary injunction. The court's decision is reported at Concord Music Group, Inc. v. Anthropic PBC, 772 F. Supp. 3d 1131 (N.D. Cal. 2025), and at 2025 WL 1482734.[25][26]
The court's reasoning addressed both prongs of the preliminary injunction standard:
Importantly, the court did not decide the fair use question on the merits. The order expressly preserved the publishers' ability to seek damages and a permanent injunction at later stages of the case.[25][27]
The day after denying the preliminary injunction, the court issued an order on Anthropic's motion to dismiss the original complaint. The court allowed the publishers' direct infringement claim to proceed but granted dismissal of the contributory infringement, vicarious infringement, and DMCA section 1202 claims, with leave to amend.[17][28]
Key holdings included:[17][28]
The publishers filed a First Amended Complaint on April 25, 2025, addressing the dismissed claims. The amended complaint provided more detail about specific user-side infringement, fleshed out the allegations against Anthropic concerning the choice of the "Newspaper" extraction library, and added new evidentiary detail about Anthropic's training pipeline drawn from materials the publishers obtained in discovery and from public filings in the parallel Bartz v. Anthropic book-author case.[9][29]
On October 6, 2025, Judge Lee denied Anthropic's motion to dismiss the First Amended Complaint. The order allowed all four counts to proceed, holding that:[4][29][30]
On October 9, 2025, Judge Lee denied the publishers' motion for leave to file a Second Amended Complaint that would have added a new claim against Anthropic for using BitTorrent to obtain copies of books containing the publishers' lyrics from shadow libraries (so-called "Bartz-style" piracy claims based on materials uncovered in the parallel Bartz v. Anthropic book-author litigation). The court reasoned that the proposed amendment came too late in the schedule and that the new claim involved sufficiently distinct facts and proof to be brought, if at all, as a separate action.[31][32]
The publishers thereafter pursued the piracy claim as a separate lawsuit.
On January 28, 2026, the same three publishers identified in press reporting as the most active in Concord I, namely Concord Music Group, Universal Music Group, and ABKCO Music, filed a new complaint in the U.S. District Court for the Northern District of California. The case is docketed as Concord Music Group, Inc. v. Anthropic PBC, No. 5:26-cv-00880, and is colloquially referred to as "Concord II."[14][33]
The Concord II complaint expanded both the universe of accused acts and the universe of accused defendants:[14][33]
On March 6, 2026, Anthropic filed a Motion to Stay in Concord II, seeking to hold that action in abeyance pending final resolution of Concord I or, alternatively, pending resolution of summary judgment in Concord I. Anthropic's stay motion was set for hearing on May 20, 2026, in front of Judge Lee.[5][34]
The Concord I case is in pretrial dispositive motion practice. According to the publicly available scheduling orders and contemporaneous reporting, all hearing and trial dates previously set in Concord I have been vacated; the parties are operating under a revised schedule in which summary judgment briefing is to be fully completed by June 8, 2026, with a hearing on dispositive motions scheduled for July 15, 2026. Commentators have widely characterized Judge Lee's eventual fair use ruling as one of the next major AI copyright decisions to follow the June 2025 decisions of Judge William Alsup in Bartz v. Anthropic and Judge Vince Chhabria in Kadrey v. Meta.[4][5][35]
Concord Music Group v. Anthropic was the first U.S. lawsuit by music publishers against a generative AI developer for the use of song lyrics, although it has since been followed by suits against music-generation AI services by the major record labels and by additional publisher-side actions. Its docket has become a leading vehicle for fair use, secondary liability, and DMCA questions specific to lyrics, which differ from the questions presented by books or news articles. Lyrics are typically short, highly memorable, and licensed in well-developed digital aggregator markets (such as MusixMatch, LyricFind, and Genius), which gives the publishers a uniquely strong argument that an AI training market should be analyzed against an active, paid licensing baseline rather than a non-existent or nascent market.[7][9]
The January 2025 stipulation is notable as one of the first court-approved agreements specifying technical guardrails on a generative AI system. The stipulation requires that Anthropic's guardrails apply to both current and future products and includes a structured notice-and-investigation procedure under court oversight. Commentators have described the stipulation as a template that other rightsholders may seek to replicate in cases against other AI developers, and as a partial illustration of how the courts may end up policing AI outputs going forward even where they decline to enjoin training-side conduct.[22][23][24]
Judge Lee's March 2025 preliminary injunction order is widely cited for the proposition that allegations of harm to a hypothetical "AI training market" are too speculative, at least at the preliminary injunction stage, to support irreparable harm in an AI training copyright case. The ruling reinforces the practical pattern emerging across AI copyright cases in which preliminary injunctions are difficult to obtain and disputes are pushed instead to merits adjudication on the fair use defense and to monetary remedies.[25][36]
Judge Lee's two motion-to-dismiss orders illustrate a recurring problem in generative AI copyright pleading: rightsholders frequently produce examples of infringing outputs that were generated by their own investigators, who do not themselves constitute infringing third parties for purposes of contributory or vicarious liability. The court's March 2025 order required the publishers to plead more concrete user-side direct infringement to maintain their secondary liability theories, while the October 2025 order found that the First Amended Complaint had cleared that bar. The arc of the rulings sets out a roadmap that other rightsholder plaintiffs and defendants will likely follow in similar cases.[17][29]
The October 2025 order is also noteworthy for its treatment of DMCA section 1202. The court held that allegations of intentional tool selection (Anthropic's alleged choice of the "Newspaper" library because it was "more effective" at stripping CMI than alternatives, combined with Anthropic's alleged knowledge that language models memorize and reproduce training data) can plausibly establish the scienter required for DMCA section 1202 liability in an AI training context, an area of law that had previously been litigated almost exclusively in cases such as Doe 1 v. GitHub and Tremblay v. OpenAI with mixed outcomes.[29][37]
The pending summary judgment briefing in Concord I, scheduled to conclude in June 2026, is widely expected to be the principal vehicle for Judge Lee's fair use ruling. Because the case presents both training-side and output-side allegations, the court will likely have to address several questions that earlier AI fair use rulings either declined to reach or addressed only partially:
Because Judge Lee has held that fair use is not amenable to resolution at the pleadings stage in this matter and because the preliminary injunction order expressly did not decide the question, the summary judgment ruling, if it reaches the merits, will be the first substantive judicial decision on these questions in the music-publishing context.[4][25]
As of May 19, 2026, the matter is in the following posture:[3][4][5]
The case has not settled. As of the same date, no public reports indicated that Concord I had been formally consolidated with Bartz v. Anthropic (the related book-author class action against Anthropic that produced its own June 2025 fair use ruling and an announced $1.5 billion settlement in late 2025), although the cases share several factual themes and an overlapping discovery record concerning Anthropic's BitTorrent activity.[4][14][35]