On August 28, Sony Music Publishing and Warner Chappell sued Anthropic — and, personally, CEO Dario Amodei and co-founder Benjamin Mann — accusing the company of a "brazen campaign" of illegally torrenting, scraping, and downloading copyrighted song lyrics to train Claude. It's the fifth music-publisher lawsuit against Anthropic in under a year, and it lands in a week that made unusually clear that AI vendor risk isn't just about price or infrastructure.
What Changed
According to the 48-page complaint, Anthropic's data-cleaning process allegedly stripped song titles, songwriter names, publisher information, and copyright notices — while keeping the lyrics themselves, making the protected material harder to identify and letting model outputs appear without attribution. The publishers are demanding a jury trial and seeking up to $150,000 per infringed work, plus $25,000 for each instance of removed copyright management information. The case adds to a pattern: Universal Music Publishing, Concord Music, and ABKCO sued Anthropic in October 2023 and filed a second suit in January 2026 covering more than 20,000 works; BMG sued in March; Round Hill Music followed on August 17. Context matters here: this kind of dispute isn't exclusive to Anthropic — OpenAI, Google, Meta, and other generative AI companies face similar suits from rights holders.
Why It Matters
There's a distinction worth making that often gets missed in AI vendor contracts: most major labs — Anthropic included — offer enterprise customers indemnification against third-party claims arising from authorized use of a model's output (the response Claude generates for your specific use case). This kind of lawsuit is different: it's a claim against the vendor's own training practices, a risk the vendor carries alone, not something the standard output-indemnification clause covers. In other words, your company likely isn't directly exposed by this particular suit — but it shows that contractual indemnification has a boundary, and it's worth knowing exactly where it sits.
The Impact for Brazil
Brazilian companies negotiating AI vendor contracts should ask two specific questions before signing: first, what exactly does the indemnification clause cover — model output only, or exposure tied to training data as well? Second, what happens to the contract if the vendor faces a major judgment or settlement — does it affect pricing, model availability, or product continuity? This isn't different in spirit from the question we raised here about data jurisdiction with Alibaba Cloud: AI vendor risk today is multidimensional, and a generic compliance clause rarely covers every angle.
Entercast's Take
This week gave an even more direct example of how relationships between AI vendors can turn into business risk: OpenAI announced, also on August 29, that it will cut off model access to Cursor starting November 12, after SpaceX acquired the tool — validating exactly the vendor-dependency risk we flagged here when that acquisition happened, on August 19. Between an intellectual-property dispute, a contractual break between labs, and a data-sovereignty question, choosing an AI vendor in 2026 has stopped being just a benchmark-driven technical decision — it's a business-risk decision, and it deserves due diligence of the same scale.