Sony Music and Warner Chappell Sue Anthropic Over Claude Training Data

The complaint does not stop at Anthropic the corporation - it names Dario Amodei and Benjamin Mann personally, alleging the company "conducted a brazen campaign of illegally torrenting, scraping, and downloading copyrighted works on a massive scale" ^[1]. That personal-liability framing sits alongside a second legal theory under DMCA Section 1202. The complaint alleges Anthropic downloaded pirate-library book collections and scraped licensed lyric databases like MusixMatch and LyricFind to build its training corpus ^[4], then stripped out copyright notices and proprietor names from the resulting works whilst leaving the expressive content intact ^[9]. Stacking that claim onto standard infringement matters financially. Statutory damages run up to $150,000 per willfully infringed work, and CMI removal adds up to $25,000 per instance, which is how a suit covering tens of thousands of songs gets described as multi-billion dollar ^[2]. Naming the founders individually also raises the stakes beyond a corporate settlement check - it tests whether AI-company leadership can be held personally accountable for data-sourcing decisions made years before a product shipped.

Sony and Warner Chappell are not the first music-rights holders to sue Anthropic, and the timeline shows an industry building its case in stages rather than opening on every AI lab at once. Universal Music Group and other publishers first sued in 2023 over roughly 500 songs' lyrics being reproduced by Claude ^[6]. In January 2026, Universal Music Group, Concord, and ABKCO filed a larger suit alleging 714 works obtained via torrents and 20,517 songs used in training or outputs ^[5], seeking more than $3 billion, using legal representation that overlaps with the Sony/Warner Chappell case ^[6][7]. That case landed just months before Anthropic's unrelated $1.5 billion Bartz v. Anthropic book-piracy settlement won final court approval in mid-2026 ^[8]. The Sony/Warner Chappell complaint, covering tens of thousands of compositions, is now described as broader in scope than those earlier actions - and one widely-shared account of the filing separately noted BMG has its own separate lawsuit against Anthropic covering a comparatively narrow set of about 493 compositions, reinforcing a reading that each successive suit tests a bigger claim against the same defendant before other labels or other AI companies follow.

Beyond the training-data allegations, the complaint makes a forward-looking argument: that Anthropic's lyric guardrails are "easily circumventable by simply re-prompting the model" ^[2]. Publishers pressed the point further after Anthropic pointed to newer guardrails installed after the suit was filed, arguing those too are "porous, allowing all forms of infringing outputs" and could plausibly be relaxed once litigation ends ^[3]. If a court credits that argument, the consequence extends past the damages calculation - it undercuts the copyright indemnity protections Anthropic offers enterprise customers, since indemnity commitments generally assume the underlying model reliably refuses to reproduce protected material rather than needing a fresh prompt each time it slips through.

Sentiment across social platforms leaned negative toward Anthropic, with mainstream outlets echoing the complaint's own framing of the conduct as one of the largest and most blatant ongoing thefts of intellectual property in history. The story broke quietly overnight with no announcement from either side, and independent accounts - not press releases - were what first drove attention to the filing before newsrooms picked it up. One general-interest tech community's discussion of the suit crossed 12,000 upvotes within a day, an unusually large crossover for a music-industry legal filing into a non-music audience. But the dominant read in comment threads was cynical rather than sympathetic: many framed this as a fight between two sets of large corporations rather than a fight on behalf of songwriters, expecting Anthropic to treat any eventual payout as a cost of doing business given the Bartz precedent. Discussion also centered on the practical evidentiary question of how plaintiffs will prove specific songs were used in training, with prompt-extraction and regurgitation attacks floated as the likely proof strategy, alongside skepticism about why the label groups have targeted Anthropic specifically rather than every major AI lab at once.