GEMA v. Suno
| Court | Munich Regional Court I |
| Country | Germany |
| Parties | Claimants: GEMA Defendant: Suno Inc |
| Date Claim Issued | 21 January 2025 |
| Type of Claim | Copyright infringement. |
| Status as of May 2026 | Oral hearing held on 9 March 2026; deadline for the defendant to submit written pleadings by 7 April 2026; judgment scheduled for 12 June 2026. |
| Summary of Key Background Facts | The claimant alleges copyright infringement in connection with an AI-based music generator operated by the defendant. It is undisputed that the defendant’s model was trained using six well-known musical works. For this purpose, the defendant used stream-ripping techniques to extract content from YouTube and circumvent technical protection measures (“rolling cipher”). To generate the disputed outputs, the claimant entered prompts consisting of the original song lyrics, the title of the work and a desired musical style. No specifications were given regarding melody, harmony, rhythm or arrangement. The claimant argues that the outputs show a strong similarity to the original works and therefore demonstrate that the works were “memorised” in the model. |
| Remedies sought | Injunctive relief, information and damages |
| Summary of key legal arguments | Claimant’s claim The claimant argues that the outputs of the music generator constitute copyright infringement and that unlawful reproductions already take place within the underlying AI model. According to the claimant, the similarity of the outputs to the original works shows that the works have been “memorised” by the model. Furthermore, the output of the generated content constitutes additional independent infringements. The alleged infringements concern both Germany and the United States. Defendant’s defence The defendant argues that the musical works at issue are allegedly not protected by copyright and are not recognizable in the outputs. The training data are neither contained in nor stored within the model; rather, the model only represents mathematical patterns and relationships derived from the training data. The outputs are primarily based on the claimant’s prompts and therefore cannot be attributed to the defendant. In addition, the training is justified under US law by the doctrine of “fair use”, and the court lacks jurisdiction to assess this. To the extent that German law applies, the defendant submits that there is no copyright-relevant act of use or that any such use is justified by statutory exceptions, in particular for text and data mining. Court’s Judgment No decision has been issued yet. The court has discussed the facts and legal issues with the parties and granted the defendant a deadline to submit written pleadings by 7 April 2026. A date for the delivery of the judgment has been set for 12 June 2026. |
