AI and copyright: Anthropic reaches legal agreement with record labels

The music industry's first victory over Anthropic: the AI provider must install guard rails. However, the question of training remains open.

listen Print view
Colorful,Audio,Waveform,On,Virtual,Human,Background,,represent,Digital,Equalizer

(Image: whiteMocca/Shutterstock.com)

3 min. read

Anthropic and several record labels have signed a court settlement in the USA. This is like a first victory for the music industry. The AI provider declares that its AI models will no longer output copyrighted lyrics and will not use them as the basis for producing similar lyrics. The original lawsuit dates back to the fall of 2023 and the issue of whether the music and lyrics may be used for training AI models remains open.

The plaintiffs include Universal Music, Concord Music and ABKCO, with whom the Rolling Stones, for example, are under contract. They have accused Anthropic of using millions of songs to train its own AI models – and to play them back at the user's request. More than 500 songs were selected as representatives for the lawsuit, including songs by Katy Perry, Beyoncé and numerous other well-known artists.

Anthropic has incorporated safeguards into future AI models to ensure that no copyrighted songs by the plaintiffs are used by the AI as output. It is also planned that the record labels will be provided with the opportunity to intervene if these safeguards do not function as intended. They can point out such cases to the AI provider, who must then react immediately.

Anthropic assures that it is unlikely that the AI models, or specifically the AI chatbot Claude, will reproduce song lyrics one-to-one. There are also already protective measures in place to prevent this from happening.

Videos by heise

A further court ruling is also expected on the use of song lyrics and music for AI training. However, this may take a few more months. In the lawsuit, the plaintiffs accuse Anthropic of "direct copyright infringement, contributory infringement, vicarious infringement and removal or alteration of copyright management information in violation of copyright law". The now-signed agreement issued by the court explicitly includes a paragraph stating that this issue has not yet been resolved. "In particular, the parties continue to dispute the publishers' request in their motion for preliminary injunction, which is neither resolved nor affected by this agreement", it states.

Anthropic, like other AI providers, has so far relied on the US principle of fair use. This allows the use of material under copyright as long as it benefits everyone.

In addition to this lawsuit, there are other lawsuits with similar allegations. GEMA, for example, is even suing OpenAI. The music collecting society has gone to the Munich Regional Court. There are other lawsuits from publishers, the most prominent example being the lawsuit from the New York Times, but smaller publishers and individual artists also want to protect their interests in court. Lawsuits are also being brought against the providers of image generators, for example the copyright lawsuit brought by Sarah Andersen against Stability.

Case number: Case 5:24-cv-03811-EKL

(emw)

Don't miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.