Music Publishers Sue Anthropic Over Allegedly Pirated Claude Training Material

01Sony Music Publishing and Warner Chappell Sue Anthropic, Alleging It Pirated Works to Train Claude

Sony Music Publishing, Warner Chappell and other music publishers have sued Anthropic, the AI laboratory behind the Claude model family, alleging it illegally obtained copyrighted material for training. The complaint also names Anthropic co-founders Dario Amodei and Benjamin Mann as defendants.

The lawsuit, filed late Friday in the US District Court for the Northern District of California, seeks damages for “tens of thousands” of copyrighted works. The publishers are requesting up to $150,000 per work, plus as much as $25,000 for each alleged removal of identifiable copyright information. The total could reach several billion dollars if the court rules for the publishers and awards the maximum amounts.

The case distinguishes between using copyrighted material to train an AI model and the way that material was acquired. The publishers allege Anthropic conducted a campaign of torrenting, scraping and downloading works to develop and operate Claude, rather than obtaining the content through authorized channels.

According to the complaint, Mann used BitTorrent—a peer-to-peer file-sharing system—to download more than five million pirated books. Anthropic employees allegedly downloaded at least another two million books from Pirate Library Mirror. The publishers say some of those books contained lyrics and sheet music.

The complaint further alleges that Anthropic scraped song lyrics from services including MusixMatch and LyricFind, which had paid to license the publishers’ content. Songs allegedly found in Anthropic’s training data include “Ain’t No Mountain High Enough,” “Livin’ On a Prayer,” “September,” “Hallelujah” and “Paper Rings.”

That focus on acquisition connects the new lawsuit to Bartz v. Anthropic, a case brought by authors over copyrighted works used to train products including Claude. In that case, a judge found that using copyrighted works for AI training was lawful but acquiring them through piracy was not. Anthropic was ordered to pay $1.5 billion.

The music publishers are now advancing a similar argument on a broader set of alleged downloads and scraped lyrics. Anthropic disputes their account. “We disagree with the publishers’ claims and we intend to defend ourselves robustly in court,” a company spokesperson told TechCrunch.

No court has yet determined whether the alleged torrenting, downloading, scraping or removal of copyright information occurred, and no damages have been awarded in this case.

02Tracer.AI Copyright Notice Filed for Microsoft Takes Luanti Off Google Play

Luanti’s Android app is no longer available through Google Play after Tracer.AI filed a copyright notice on Microsoft’s behalf alleging infringement of Minecraft assets. The removal immediately cuts off the store as a download channel for users of Luanti, a nonprofit, community-developed, open-source platform for creating and playing block-based games.

The notice was submitted under the US Digital Millennium Copyright Act, whose notice-and-takedown process prompts distributors such as Google Play to act on reported infringement. According to Luanti, the complaint says its products use copyrighted assets taken directly from Minecraft without authorization and cites US registration TX 8-192-097, covering Minecraft Java Edition 1.9.

Luanti says the notice does not identify any particular texture, image, sound, code or other asset that allegedly infringes that copyright. The project calls the claim baseless, but the source provides no response from Google, Microsoft or Tracer.AI.

The project’s rebuttal begins with how Luanti is distributed. It says the app is a game-creation platform that ships without games or game assets by default. Users instead download community-made games and other material or join multiplayer servers. Luanti also says its included fonts and other assets are properly attributed.

Its repository contains a Development Test game used to test the engine, but Luanti says that content is no longer included in official releases. Minetest Game, a basic crafting-and-survival game once bundled with the platform, has likewise been a separate download since December 2023; the project says its assets are original and properly licensed.

Third-party games, modifications and texture packs are available through ContentDB, Luanti’s in-app catalog. The project says volunteer staff manually review uploaded packages, including checks for commercial or copyrighted material. It has also considered conventional perceptual hashing, which compares visual similarities, to flag possible problems for review, while retaining a human decision rather than relying on automated findings.

Luanti says Tracer.AI sent a similar notice in 2023 and that the project successfully appealed it. For the current removal, however, the source gives no date for Google to complete any review and does not establish whether or when the Android app will return to the store.

03Debian Votes Neither to Ban Nor Endorse Generative AI: Contributors Must Still Understand, Review and Test Outputs

Debian has adopted a position on the “responsible use of generative AI,” setting common expectations for AI-assisted work across the volunteer-run software project. The decision covers the development and maintenance of software, software packaging, documentation, and other media published within Debian.

The project neither endorses nor prohibits using generative AI tools for that work. Instead, it treats AI as one possible tool that may improve contributor productivity, while making clear that its use does not lower Debian’s standards or amount to project approval of AI-generated material.

Debian said responsible use could allow volunteers to spend more of their limited time on tasks requiring technical expertise, judgment, review, and collaboration. That potential benefit, however, does not create a separate or more permissive route for AI-assisted submissions.

All contributions must meet the same requirements regardless of how they were produced. Debian expects submitted work to satisfy its standards for quality, correctness, maintainability, and legal compliance, whether a contributor used generative AI, another tool, or no such assistance.

In practice, contributors cannot simply pass an AI tool’s output into the project without taking ownership of it. They are expected to understand the output, review it, test it, and modify it where appropriate before incorporating it into Debian. The policy therefore places responsibility on the person submitting the work, rather than on the tool that helped produce it.

That allocation of responsibility also applies when an AI-assisted contribution contains errors or creates legal concerns. Debian’s position says using generative AI does not diminish the contributor’s responsibility for submitted work. The decision establishes a tool-neutral rule: the method used to create a contribution does not change who must ensure that it is suitable for the project.

The available source does not give the vote count or describe how Debian will identify violations, enforce the position, or resolve disputes over AI-assisted work.