A coalition of music publishers led by Harmony Music Group and Common Music Group filed a federal lawsuit on January 28, 2026, alleging Anthropic illegally downloaded greater than 20,000 copyrighted musical compositions utilizing BitTorrent file-sharing know-how from infamous pirate web sites. The criticism seeks damages that would exceed $3 billion.

The lawsuit names Anthropic PBC, CEO Dario Amodei, and co-founder Benjamin Mann as defendants. Filed in the USA District Courtroom for the Northern District of California, the criticism alleges the AI firm used BitTorrent to accumulate thousands and thousands of pirated books from Library Genesis and Pirate Library Mirror, together with tons of of songbooks and sheet music collections containing copyrighted lyrics owned by the publishers.

“Whereas Anthropic misleadingly claims to be an AI ‘security and analysis’ firm, its document of unlawful torrenting of copyrighted works makes clear that its multibillion-dollar enterprise empire has in reality been constructed on piracy,” the criticism states.

Separate claims from earlier lawsuit

This represents the second main copyright motion the music publishers have introduced in opposition to Anthropic. The plaintiffs beforehand sued the corporate in October 2023 over its unauthorized use of 499 musical compositions in coaching and output from Claude AI fashions, a case often known as Harmony Music Group v. Anthropic PBC.

The publishers tried to amend their unique criticism to deal with the newly found torrenting violations after Decide William Alsup revealed Anthropic’s BitTorrent actions in a July 2025 ruling within the separate Bartz v. Anthropic case. Anthropic efficiently opposed that modification, arguing the torrenting claims have been “fully unrelated” to the unique lawsuit and would “basically remodel” the case.

The present lawsuit addresses two distinct classes of alleged infringement. First, it covers Anthropic’s downloading and distributing of copyrighted works through BitTorrent from pirate libraries. Second, it alleges ongoing copying of publishers’ works in coaching and output from Claude AI fashions launched after the publishers filed their amended criticism within the first case.

Anthropic launched a number of new Claude variations since that preliminary submitting, together with Claude 4.5 Sonnet on September 29, 2025, Claude 4.5 Haiku on October 15, 2025, and Claude 4.5 Opus on November 24, 2025. The criticism alleges every of those fashions was skilled utilizing unauthorized copies of the publishers’ musical compositions.

BitTorrent downloads from pirate websites

The lawsuit particulars how Anthropic executives, together with Amodei and Mann, personally mentioned and licensed the unlawful downloading of thousands and thousands of books from Library Genesis and Pirate Library Mirror utilizing BitTorrent protocols.

In June 2021, Mann personally used BitTorrent to obtain roughly 5 million copies of pirated books from LibGen, in accordance with the criticism. Earlier than continuing, he mentioned the plan with Amodei, chief science officer Jared Kaplan, and different senior management. Anthropic’s Archive Workforce had deemed LibGen a “blatant violation of copyright,” whereas Amodei himself described the pirate library as “sketchy.”

Regardless of these issues, firm management permitted the torrenting exercise. Amodei admitted on the time that Anthropic “ha[d] many locations from which” it might have legally bought these copyrighted works for coaching however selected to illegally torrent them as an alternative as a result of it was sooner and free, the criticism states. In Amodei’s personal phrases, they did so to keep away from a “authorized/apply/enterprise slog.”

In July 2022, Anthropic downloaded thousands and thousands of further pirated books from Pirate Library Mirror, a shadow library that mirrored the contents of the shuttered Z-Library. When one Anthropic founder found he might torrent further works from PiLiMi, he messaged colleagues “[J]ust in time!” One other worker responded, “zlibrary my beloved,” in accordance with the criticism.

The catalogs of LibGen and PiLiMi bibliographic metadata reveal that among the many thousands and thousands of books Anthropic downloaded have been tons of containing sheet music and track lyrics to musical compositions owned by the publishers. These included giant numbers of books revealed by the publishers’ sheet music licensees Hal Leonard and Alfred Music.

Particular titles talked about within the criticism embody The Finest Songs Ever that includes “Candle within the Wind” and “Each Breath You Take,” VH1’s 100 Biggest Songs of Rock & Roll that includes “All Alongside the Watchtower” and “Good Vibrations,” Rolling Stones – Let It Bleed: Genuine Guitar TAB that includes “Gimme Shelter,” Elton John – Biggest Hits Songbook that includes “Rocket Man,” Creedence Clearwater Revival: Simple Guitar that includes “Have You Ever Seen The Rain,” and Harry Kinds Songbook that includes “Signal of the Instances.”

Two-way infringement through BitTorrent

The BitTorrent protocol’s peer-to-peer nature creates simultaneous importing and downloading. When Anthropic downloaded copies of pirated books through torrenting, the protocol concurrently uploaded to the general public unauthorized copies of the identical works.

This two-way copying violated each the publishers’ unique proper of copy by means of downloading and their unique proper of distribution by means of importing. The criticism alleges every pirated work Anthropic torrented was doubtless shared hundreds or tens of hundreds of occasions, depriving publishers of considerable income.

“Defendants’ use of BitTorrent triggered intensive hurt to Publishers,” the criticism states. “Every pirated work Defendants torrented was doubtless shared hundreds if not tens of hundreds of occasions, depriving Publishers of considerable income. Defendants additionally contributed to the continued viability of BitTorrent and pirate libraries as instruments for infringement that solely exist so long as they’ve customers.”

Anthropic copied these books to amass a “huge central library” of written texts the corporate meant to take care of without end, in accordance with the criticism. Whereas Anthropic has claimed it didn’t use any of those illegally torrented books to coach industrial Claude fashions, the lawsuit argues the torrenting itself constitutes standalone copyright infringement no matter subsequent use.

“No matter Anthropic’s later use, its piracy of those books through BitTorrent was unquestionably infringing,” the criticism states. “Even when some subset of the books Defendants illegally torrented have been typically used for AI coaching, that can’t excuse their mass torrenting of thousands and thousands of pirated books with out paying for them – together with books containing Publishers’ musical compositions.”

Continued AI coaching infringement

Past the torrenting claims, the lawsuit alleges Anthropic continues to repeat publishers’ works on a large scale for AI coaching even after the primary lawsuit was filed. The criticism identifies 20,517 musical compositions that Anthropic allegedly infringed by means of coaching newer Claude fashions and the output these fashions generate.

Anthropic trains every new Claude mannequin from scratch utilizing newly copied coaching corpora that embody the publishers’ copyrighted works, in accordance with the criticism. The corporate copies this coaching information from a number of sources, together with scraping web sites, scanning bodily books, and exploiting third-party datasets.

These datasets embody The Pile, which includes the Books3 assortment of pirated books and YouTube Subtitles containing closed captions from movies that seize lyrics to publishers’ compositions. Anthropic additionally exploits the Widespread Crawl dataset, which accommodates publishers’ copyrighted lyrics scraped with out permission from web sites of the publishers’ licensees together with MusixMatch, LyricFind, and Genius.

The criticism particulars Anthropic’s “cleansing” course of throughout coaching, which removes materials inconsistent with its enterprise mannequin however conspicuously doesn’t take away unauthorized copyrighted content material equivalent to publishers’ lyrics. As a substitute, Anthropic makes use of extractor instruments to take away copyright notices and different copyright administration data from the copied textual content.

“Anthropic desires to coach its Claude AI fashions particularly on the content material of Publishers’ musical compositions, together with the lyrics, in order that the fashions’ output will reproduce that expressive content material, slightly than copyright notices or different Copyright Administration Info accompanying these lyrics, data that’s vital to defending Publishers’ rights however which Anthropic deemed ineffective,” the criticism states.

The lawsuit features a separate declare underneath Part 1202 of the Copyright Act for elimination or alteration of copyright administration data. Music titles, creator names, and copyright proprietor data represent legally protected copyright administration data.

Anthropic has deliberately eliminated this data each throughout AI coaching and in mannequin outputs, in accordance with the criticism. As early as Might 2021, high-ranking Anthropic workers together with founders Mann and Kaplan mentioned extraction instruments used to filter coaching information.

In June 2021, Mann and Kaplan concluded that the jusText extraction software left an excessive amount of “ineffective junk” – equivalent to copyright discover data contained in footers – in scraped internet information when in comparison with different instruments. Mann expressed his want that the AI “mannequin will be taught to disregard the boilerplate,” like copyright notices.

In a single inner chat, an Anthropic workers member shared an instance displaying that when jusText was utilized to a scraped webpage containing footnotes, a copyright proprietor title, and “© 2019” copyright discover, it left that data untouched. In distinction, the Newspaper extraction software eliminated the footnotes, copyright proprietor title, and copyright discover fully, which was thought of “a major enchancment.”

“As a result of Newspaper eliminated Copyright Administration Info extra successfully, Anthropic purposefully determined to make use of that software to take away copyright notices and different Copyright Administration Info from Publishers’ lyrics and different copyrighted works,” the criticism states.

Anthropic intentionally extracts this data to stop its fashions from displaying copyright notices alongside publishers’ lyrics in outputs, thereby concealing the corporate’s infringement from customers, publishers, and different copyright homeowners, in accordance with the criticism.

Claude fashions memorize and regurgitate lyrics

The lawsuit alleges Anthropic’s Claude fashions are designed to “memorize” and regurgitate their coaching information, together with publishers’ copyrighted lyrics. This tendency is effectively documented and well-known to Anthropic, in accordance with the criticism.

In July 2020, a number of synthetic intelligence researchers at OpenAI – together with future Anthropic founders Amodei, Mann, Jack Clark and Kaplan – noticed that “a serious methodological concern with language fashions pretrained on a broad swath of web information, significantly giant fashions with the capability to memorize huge quantities of content material, is potential contamination of downstream duties by having their check or growth units inadvertently seen throughout pre-training.”

An Anthropic inner report said extra bluntly: “Massive LMs memorize A LOT, like a LOT,” in accordance with the criticism.

The lawsuit argues these regurgitations are a characteristic slightly than a bug. Anthropic understands that Claude customers particularly search publishers’ lyrics and derivatives of these lyrics, and it has developed and skilled Claude to reply to exactly these sorts of requests.

When growing Claude’s fine-tuning course of, Anthropic employed short-term employees to speak with the AI. In written directions, Anthropic offered instance duties together with “suggesting songs primarily based in your favourite music” or “ask[ing] fashions to re-write textual content with model, content material, and formatting adjustments or requests.”

Anthropic’s personal workers continuously prompted Claude for track lyrics and derivatives when growing, testing, and utilizing Claude fashions, in accordance with the criticism. In January and February 2023, shortly earlier than first releasing Claude to the general public, quite a few Anthropic workers mentioned prompting Claude for copies of publishers’ lyrics.

Anthropic founder and chief compute officer Tom Brown queried “@Claude what are the lyrics to desolation row by [Bob] Dylan?” One other worker prompted the mannequin to “write a coherent poem made up of fragments” of “lyrics from the Beatles, Bob Dylan, and different classics from the 60s/70s.” A 3rd worker requested Claude “What are the lyrics to we discovered love by Calvin Harris?”

The criticism alleges that after Anthropic launched its fashions to the general public, third-party customers made related requests for publishers’ lyrics, and Claude generated responses reproducing these lyrics in violation of publishers’ rights.

Insufficient guardrails

After publishers filed their first lawsuit and publicly uncovered Anthropic’s infringement, the corporate adopted further guardrails purportedly designed to reduce AI output copying publishers’ copyrighted works. The criticism alleges these guardrails stay insufficient.

Anthropic intentionally selected to incorporate lyrics for less than a restricted variety of particular songs as a part of its guardrails, together with the five hundred works recognized within the first lawsuit. The guardrails is not going to comprehensively forestall output copying lyrics from the a lot broader universe of copyrighted songs past that restricted set, in accordance with the criticism.

The guardrails are usually not designed to dam all prompts and output that will copy or include publishers’ copyrighted works, equivalent to requests that Claude generate supposedly “new” or “unique” songs within the model of particular artists. Anthropic’s fashions proceed to generate output containing publishers’ lyrics even when not particularly requested.

Scientific literature confirms Claude will nonetheless ship giant quantities of copyrighted content material as output regardless of the guardrails, in accordance with the criticism. Recent research published in January 2026 confirmed Stanford researchers efficiently extracted giant parts of copyrighted books from Claude 3.7 Sonnet, with the mannequin reproducing 95.8% of Harry Potter and the Sorcerer’s Stone practically verbatim.

“What’s extra, as a result of these guardrails handle solely Claude output, and do nothing to stop Anthropic’s underlying exploitation of Publishers’ lyrics in AI coaching, they’re at most a band-aid – not a treatment – for Anthropic’s infringement,” the criticism states.

Monetary hurt to publishers

The lawsuit alleges Anthropic’s conduct has triggered substantial and irreparable hurt to publishers and their songwriters. Anthropic’s use of publishers’ works with out licenses deprives publishers of license charges and undercuts all the licensing market.

Anthropic has created a software, skilled on unauthorized copies of publishers’ works, that allows customers to generate huge portions of AI-generated lyrics and songs that compete with publishers’ reliable copyrighted works. This competitors harms the marketplace for and worth of these works.

Distribution by means of BitTorrent is especially pernicious, in accordance with the criticism, as a result of every file could be distributed tons of or hundreds of occasions by means of the swarm. Anthropic’s widescale use of BitTorrent contributes to the continued viability and normalization of that infringing protocol.

The elimination of copyright administration data makes it tougher for publishers to implement their copyrights and shield their works from additional exploitation. The sheer breadth and scope of Anthropic’s copying makes it “successfully inconceivable to measure, calculate, and even estimate the monetary harm it imposes on songwriters and publishers,” the criticism states.

Anthropic is now valued at $350 billion or extra, up from $183 billion in September 2025. The corporate has acquired billions of {dollars} in funding from Amazon, Google, and different traders. Anthropic nearly doubled its valuation in just two months from September to November 2025.

The criticism alleges one of many important causes Anthropic’s AI fashions are common and beneficial is as a result of the corporate skilled these fashions on a textual content corpus that features publishers’ copyrighted lyrics. Publishers’ copyrighted content material serves as a draw for particular person customers, industrial clients, and traders.

Willful infringement allegations

The lawsuit alleges Anthropic’s infringement is willful, intentional, and purposeful, in disregard of and with indifference to publishers’ rights. Anthropic is aware of it’s utilizing plaintiffs’ works with out permission, but has skilled and publicly launched a number of new variations of Claude since publishers filed their first lawsuit.

Anthropic intently screens and analyzes consumer interactions with Claude and the output generated by Claude. The corporate collects consumer prompts and corresponding outputs to check particular methods Claude is getting used. Anthropic is effectively conscious primarily based on this examine that customers request lyrics to publishers’ works and that Claude delivers copies of these lyrics.

Evaluation of Claude utilization information has revealed clusters of requests to “assist me discover, analyze, or modify track lyrics,” “translate songs or lyrics between languages,” and “assist me determine or discover songs with particular traits,” in accordance with the criticism. These clusters embody Claude prompts and output referring to publishers’ lyrics particularly.

Anthropic carried out guardrails after the primary lawsuit as a result of it knew its fashions had been skilled on publishers’ lyrics, it monitored Claude consumer exercise and output, it understood customers have been prompting fashions concerning publishers’ lyrics, and it knew the fashions have been producing particular output that unlawfully copied publishers’ lyrics.

“When Anthropic first developed and later refined and expanded these guardrails, and when it monitored the effectiveness of the guardrails, it collected and analyzed Claude prompts and output information, together with particular infringing output copying copyrighted works,” the criticism states.

This lawsuit arrives amid intensifying copyright litigation throughout the AI trade. Anthropic agreed to a $1.5 billion settlement in September 2025 to resolve copyright infringement claims from authors over its use of pirated books to coach Claude fashions.

That settlement within the Bartz case grew to become the biggest publicly reported copyright restoration in historical past. The case centered on roughly 500,000 revealed works, with authors receiving roughly $3,000 per work.

Judge William Alsup delivered a landmark split decision in June 2025 ruling that utilizing copyrighted books to coach giant language fashions constitutes transformative honest use underneath copyright regulation. Nonetheless, he allowed claims over pirated content material to proceed to trial, discovering that Anthropic’s technique of buying content material by means of piracy was not protected.

“The decide understood the outrageous piracy,” Authors’ Guild CEO Mary Rasenberger mentioned on the time. “The piracy legal responsibility comes with statutory damages for intentional copyright infringement, that are fairly excessive per e-book.”

The music publishers reached a partial agreement with Anthropic in January 2025 concerning copyright safety measures of their first case. That stipulation required Anthropic to take care of present “Guardrails” designed to stop copyright-infringing outputs and established a notification course of for publishers to alert Anthropic when guardrails fail.

Nonetheless, that settlement addressed just one portion of the publishers’ preliminary injunction movement and didn’t resolve disputes concerning the usage of copyrighted lyrics in coaching future AI fashions.

Business implications

The lawsuit highlights tensions throughout the AI trade concerning content material licensing and honest use. Publishers within the criticism notice they’ve already begun to discover and enter into licenses allowing licensed makes use of of their musical compositions in reference to AI.

Common Music Publishing Group lately entered into agreements with AI music generator Udio and AI music know-how firm KLAY to license sure works in reference to AI coaching, in accordance with the criticism. Different giant music publishers have equally licensed AI firms to make use of their works.

“Publishers acknowledge the nice potential of moral AI as a strong software for the long run, and have already begun to discover and enter into licenses allowing licensed makes use of of their musical compositions in reference to AI,” the criticism states. “Nonetheless, it stays essential that AI know-how be developed and employed ethically and responsibly, in a way that protects the rights of Publishers and songwriters, their livelihoods, and the inventive ecosystem as an entire.”

This case arrives as a number of AI firms face related copyright challenges. Ziff Davis filed a major lawsuit against OpenAI in April 2025 accusing the corporate of unauthorized use of content material from 45 properties together with CNET, IGN, and Mashable. Reddit sued Anthropic in June 2025 over alleged unauthorized AI coaching on platform information.

The Copyright Workplace launched main steerage in Might 2025 addressing when AI builders want permission to make use of copyrighted works. The report prompt that transformativeness and market results could be probably the most important elements in honest use determinations.

Congress introduced the TRAIN Act in July 2025, which might grant copyright homeowners subpoena energy to determine works utilized in generative AI coaching with out their permission or compensation.

The publishers search statutory damages underneath the Copyright Act, which gives for damages as much as $150,000 per work for willful infringement. With 20,517 works recognized in Exhibit B alone, statutory damages might theoretically attain into the billions of {dollars}.

The criticism additionally seeks injunctive aid requiring Anthropic and its officers, together with Amodei and Mann, to stop infringing publishers’ copyrights. Publishers request an order requiring Anthropic to offer an accounting of coaching information, coaching strategies, and recognized capabilities of its AI fashions.

Moreover, publishers search an order requiring defendants to destroy underneath courtroom supervision all infringing copies of publishers’ copyrighted works in defendants’ possession or management.

The plaintiffs are represented by intensive authorized groups together with Oppenheim + Zebrak LLP, Coblentz Patch Duffy & Bass LLP, and Cowan, Liebowitz & Latman P.C. The criticism was filed in the USA District Courtroom for the Northern District of California and calls for a jury trial.

Anthropic didn’t reply to requests for remark.

Timeline

Abstract

Who: Harmony Music Group, Common Music Group, ABKCO Music, and different main music publishers filed the lawsuit in opposition to Anthropic PBC, CEO Dario Amodei, and co-founder Benjamin Mann in the USA District Courtroom for the Northern District of California.

What: The publishers allege Anthropic illegally downloaded greater than 20,000 copyrighted musical compositions utilizing BitTorrent from pirate web sites Library Genesis and Pirate Library Mirror, together with tons of of songbooks and sheet music collections. The lawsuit additionally alleges ongoing copyright infringement by means of coaching of newer Claude AI fashions and the outputs these fashions generate. Claims embody direct copyright infringement by means of torrenting, ongoing infringement in AI coaching and output, contributory and vicarious infringement by means of Claude customers, and elimination of copyright administration data.

When: The lawsuit was filed on January 28, 2026. The alleged torrenting occurred primarily in June 2021 and July 2022. The continued AI coaching infringement includes Claude fashions launched from September 29, 2025, by means of November 24, 2025, and fashions presently in coaching.

The place: The case was filed in the USA District Courtroom for the Northern District of California. Anthropic is headquartered in San Francisco. The alleged BitTorrent downloads occurred from abroad pirate library web sites Library Genesis and Pirate Library Mirror. The copyrighted works embody musical compositions owned by publishers with operations in Nashville, Santa Monica, New York, London, and Stockholm.

Why: This lawsuit issues for the advertising and marketing neighborhood as a result of it represents one of many largest non-class motion copyright instances in U.S. historical past, with potential damages exceeding $3 billion. The case addresses basic questions on how AI firms can legally purchase coaching information and whether or not present guardrails adequately shield copyrighted content material. The end result will affect licensing markets for inventive content material, have an effect on the viability of AI-powered advertising and marketing instruments, and doubtlessly reshape how AI firms method content material acquisition. As AI methods turn out to be extra deeply built-in into advertising and marketing workflows, precedents established by this case will decide whether or not firms can depend on AI platforms that skilled on doubtlessly infringing content material and what authorized dangers they face when utilizing such instruments.


Share this text


The hyperlink has been copied!




Source link