Anthropic convinced a California federal judge to reject a preliminary bid to block it from using lyrics owned by UMG and other music publishers to train Claude
Artificial intelligence company Anthropic convinced a California federal judge on Tuesday to reject a preliminary bid to block …
Context & Ripple Effects
The publishers’ dispute began with a 2023 lawsuit alleging that Anthropic used song lyrics without authorization. Before this ruling, Anthropic had separately agreed to restrict Claude from supplying song lyrics or generating new lyrics based on copyrighted material pending a court decision, creating a practical distinction between model outputs and training inputs: the interim limits on Claude’s lyric outputs.
The decision is an early procedural win in a broader copyright fight over AI training. Later coverage of Anthropic’s book-related litigation shows that courts can treat training use and the handling of pirated source libraries differently, while author claims also expanded into a potential nationwide class action.
First-order effects
- Anthropic can continue using the disputed lyrics for Claude training while the case proceeds, rather than facing the requested early injunction.
- UMG and the other publishers retain their underlying claims, but lose immediate leverage to halt the challenged training activity; Claude’s separate lyric-output restrictions remain the relevant near-term guardrail.
Second-order effects
- The ruling gives AI developers a procedural reference point against early training-data injunctions, while encouraging rightsholders to focus on evidence of infringement, outputs, and data provenance.
- Music publishers may place greater weight on controlling consumer-facing lyric generation and on licensing or litigation strategies that address training and output uses separately.
Third-order effects
- Copyright disputes over generative AI are likely to develop as a two-layer market: courts may assess model training and reproduced outputs under different standards, increasing the value of traceable, governed training corpora.
- If this pattern holds, early injunctions will be harder to obtain than remedies tied to specific unlawful outputs or improperly acquired source collections, leaving final liability questions to longer litigation.
The trend: Generative-AI copyright litigation is increasingly separating the legality of training data from the conduct of models in producing protected content.