Regardless of the jurisdiction — Sony and Warner sue Anthropic, and Recital 106 does not care where the torrenting happened

Sony Music Publishing and Warner Chappell affiliates filed against Anthropic PBC, Dario Amodei and Benjamin Mann on 28 August 2026 in the Northern District of California, No. 5:26-cv-09217, pleading copyright infringement and CMI removal under 17 U.S.C. 1202 with statutory exposure of up to $150,000 per willfully infringed work — about fifty times the roughly $3,000 a title paid under the Bartz settlement approved on 22 July 2026. The coverage framing of a ban on AI training is wrong on the case's own terms: Alsup held training on lawfully acquired books was fair use and only the pirated central library was not, and this complaint is about acquisition — five million books torrented from LibGen in June 2021, two million from PiLiMi in July 2022, lyrics scraped from MusixMatch and LyricFind. The practitioner consequence is European and easy to miss: Recital 106 requires any provider placing a general-purpose AI model on the Union market to hold a copyright policy regardless of the jurisdiction in which the copyright-relevant acts underpinning training take place, Article 53(1)(c) is that policy and 53(1)(d) the public training-content summary on the AI Office's 24 July 2025 template, Article 53(2) never lifts either even for open weights, and since 2 August 2026 the Commission may request the documentation and fine up to EUR 15 million or 3% of worldwide turnover — so the moves are to pull the 53(1)(d) summary for every model in the stack, demand the 53(1)(c) policy in writing, put provenance and indemnity in the contract, and diary 2 August 2027 for legacy models.

Post

Sony and Warner's publishers sued Anthropic on Friday. The part that matters if you ship on Claude in Europe is not in the complaint.

On 28 August, Sony Music Publishing and Warner Chappell affiliates filed in the Northern District of California — 5:26-cv-09217 — against Anthropic PBC, Dario Amodei and Benjamin Mann personally. The opening line calls it "one of the largest and most blatant ongoing thefts of intellectual property in history."

Two counts: copyright infringement, and removal of copyright management information under 17 U.S.C. 1202. Exposure runs to $150,000 per willfully infringed work and $25,000 per CMI violation. The Bartz settlement, approved 22 July 2026, paid roughly $3,000 a title.

Three things before you file this under "courts are banning AI training".

  1. It is not about training. In Bartz, Judge Alsup held training on lawfully acquired books was quintessentially transformative and fair use. What was not was assembling a permanent central library out of pirated copies. This complaint is about acquisition — roughly five million books torrented from LibGen in June 2021, two million more from PiLiMi in July 2022, lyrics scraped from MusixMatch and LyricFind.

  2. It is American litigation about American acts. Which is why the European half gets missed.

  3. Recital 106 of the AI Act: a provider placing a general-purpose AI model on the Union market must have a copyright policy "regardless of the jurisdiction in which the copyright-relevant acts underpinning the training of those general-purpose AI models take place."

Article 53(1)(c) is that policy. Article 53(1)(d) is the public summary of training content, on the AI Office's template. Both apply since 2 August 2025 — and since 2 August 2026 the Commission can request documentation, evaluate models, and fine up to EUR 15m or 3% of worldwide turnover.

Monday:

  1. Pull the 53(1)(d) training-content summary for every model in your stack. File it beside the model card.
  2. Ask for the 53(1)(c) copyright policy in writing. Article 53(2) lifts (a) and (b) for open-weight models. It never lifts (c) or (d).
  3. Put provenance and indemnity in the contract, not the incident review.
  4. Diary 2 August 2027: models on the market before 2 August 2025 have until then.

A US court will spend years deciding what was taken. The Act only ever asked the provider what it can show.

What does your model's training summary actually say?

#AIAct #AgenticAI #EUCompliance #AIGovernance #Copyright

First comment

Sources, primary first.

The complaint itself — Sony Music Publishing (US) LLC and affiliated publishers v. Anthropic PBC, Dario Amodei and Benjamin Mann, No. 5:26-cv-09217, N.D. Cal. (San Jose Division), filed 28 August 2026. This is where the "largest and most blatant ongoing thefts" line, the LibGen and PiLiMi paragraphs and the MusixMatch/LyricFind allegation sit: https://www.musicbusinessworldwide.com/files/2026/08/COMPLAINT-in-Sony_Music_Publishing_US_LLC_e.pdf

Article 53 of the AI Act — (1)(c) is the copyright policy and the Article 4(3) DSM reservation, (1)(d) is the public training-content summary, (2) is the open-source carve-out that lifts only (a) and (b): https://artificialintelligenceact.eu/article/53/

Recital 106, in full. The "regardless of the jurisdiction" sentence and the level-playing-field rationale behind it: https://artificialintelligenceact.eu/recital/106/

The Commission's mandatory template for the public summary of training content, published 24 July 2025 — this is the form the summary has to take: https://digital-strategy.ec.europa.eu/en/faqs/template-general-purpose-ai-model-providers-summarise-their-training-content

Final approval of the $1.5bn Bartz settlement, 22 July 2026, and Alsup's split between lawful acquisition and the pirated central library: https://www.jurist.org/news/2026/07/judge-approves-record-1-5-billion-settlement-involving-anthropic/

Coverage, for the framing I am arguing with — both good reports, both dating the filing to Saturday 29 August because that is when they published. The docket says Friday the 28th: https://techcrunch.com/2026/08/29/sony-music-warner-sue-anthropic-alleging-a-brazen-campaign-of-intellectual-property-theft/ https://www.axios.com/2026/08/29/anthropic-sony-warner-music-copyright

One point cut for length: the five-million and two-million book figures in this complaint are not new discovery. They come out of Bartz. One case's discovery record is now the evidentiary spine of the next one against the same defendant — which is the real reason a settlement is not the end of anything, and the reason a US settlement is not an EU compliance record either.

Claims

Sources

Corrections

None.

AI disclosure

The video for this edition contains AI-generated media: the narration voice, the cover art and the brand ident animation are generated. The script, the claims and the source verification are human-authored and human-checked. Assets carry IPTC DigitalSourceType and C2PA Content Credentials.