.png)
January 3, 2025
The AI chatbot Claude, developed by Anthropic, can no longer quote song lyrics. A settlement with major music publishers ends a legal dispute and prohibits Claude from reproducing copyrighted lyrics. This step, however, raises the fundamental question of whether AI companies are allowed to use copyrighted texts to train their models at all.
Court documents reveal that Anthropic reached an agreement with prominent companies such as the Universal Music Group, the Concord Music Group, and ABKCO. The agreement follows a lawsuit filed by the publishers in 2023, in which they accused Anthropic of allowing Claude to reproduce lyrics from at least 500 songs by various artists, ranging from Katy Perry and the Rolling Stones to Beyoncé.
Under the new agreement, Claude is not permitted to reproduce copyrighted lyrics or create new texts based on protected material. Anthropic emphasized that Claude was never intended to infringe on copyrights and that safeguards were already in place. The agreement reinforces these existing priorities. There is now also a new notification system: if publishers find that Claude is violating the rules, they can notify Anthropic in writing. The company promises to process these reports quickly.
While this agreement resolves the immediate dispute over Claude's text output, it bypasses the central question: are AI companies allowed to use copyrighted texts to train their models in the first place?
Anthropic, like other AI labs, argues that this falls under the "fair use" principle, but courts have not yet made a final decision. Anthropic states: "We continue to look forward to demonstrating that the use of potentially copyrighted material in training generative AI models, consistent with existing copyright law, represents a classic fair use."
Similar disputes are also taking place in Germany. GEMA, the German collecting society for music rights, filed a lawsuit against OpenAI in November 2024. Like the US publishers, it claims that OpenAI's ChatGPT uses song lyrics without appropriate licensing or compensation. GEMA hopes that the case will set a precedent for how AI companies should compensate rights holders.
OpenAI attempted to address these issues in May 2024 with the announcement of a "Media Manager." This tool is intended to give rights holders the ability to decide whether their works can be used for AI training or not – similar to YouTube's opt-out system for video content. However, almost eight months later, the tool is still in development, and there are no updates regarding its release.
The debate surrounding copyright and AI training thus remains open. The agreement between Anthropic and the music publishers is an important step, but the fundamental question regarding the legal use of copyrighted data for AI training still needs to be clarified. Upcoming court decisions and the further development of tools like the "Media Manager" will significantly influence the future of copyright in the age of artificial intelligence.
Bibliographie: - https://www.reddit.com/r/ClaudeAI/comments/1d9l4qz/this_is_why_you_are_getting_false_copyright/ - https://www.hollywoodreporter.com/business/business-news/anthropic-enforce-copyright-guardrails-ai-tools-1236098152/ - https://m.economictimes.com/tech/technology/anthropic-fires-back-at-music-publishers-ai-copyright-lawsuit/articleshow/106943849.cms - https://www.youreverydayai.com/3-reasons-businesses-shouldnt-use-anthropics-claude-yet/ - https://www.aiforhumans.show/anthropics-claude-35-reigns-as-openai-slips-ai-music-lawsuits-more-ai-news/ - https://podcasts.nu/avsnitt/ai-for-humans-artificial-intelligence-made-fun/anthropics-claude-35-reigns-as-openai-slips-ai-music-lawsuits-more-ai-news-AHX7tAVpEs - https://www.reuters.com/legal/litigation/anthropic-fires-back-music-publishers-ai-copyright-lawsuit-2024-01-17/ - https://www.tomsguide.com/ai/anthropic-just-gave-claude-ai-access-to-your-computer-heres-why-this-is-a-big-deal - https://medium.com/@brynn_30189/claude-ai-the-dark-horse-of-the-industry-43b8877bfa6d