Background
Independent recording artists and artist-owned entities sued Suno, the operator of a generative-AI music service. They allege that Suno downloaded their songs from YouTube, copied the works to train its models, and produced music that reproduces or closely resembles protected elements of their compositions and sound recordings.
The amended complaint asserts claims for unauthorized reproduction and derivative works under the Copyright Act, circumvention of a technological access control under section 1201(a)(1) of the Digital Millennium Copyright Act, and violation of the Tennessee Consumer Protection Act. Suno moved to dismiss the derivative-work, DMCA, and Tennessee counts; the direct-reproduction claim was not part of the motion.
The Court’s Holding
Judge F. Dennis Saylor IV allowed the derivative-work claim to proceed. Although the complaint did not identify a particular Suno output that was substantially similar to one of the plaintiffs’ songs, the court rejected a categorical rule requiring the allegedly infringing output to be attached or identified at the pleading stage. The complaint alleged actual copying for model training, about 100 examples in which Suno outputs replicated or closely resembled other copyrighted songs, and statements indicating that the system could reproduce artists’ vocals and songs. Taken together, those facts supported a plausible inference that the model had also generated outputs substantially similar to the plaintiffs’ works. Whether evidence ultimately proves that theory is for discovery and later stages of the case.
The court also allowed the DMCA claim to proceed. Plaintiffs allege that YouTube uses a “rolling cipher” to control external access and that Suno used stream-ripping tools to bypass it and download their songs. Following a related ruling in the major record labels’ case against Suno, the court held that those allegations plausibly describe circumvention of a technological measure that effectively controls access. The technical facts about the cipher and the alleged circumvention require a developed evidentiary record.
The Tennessee statutory claim was dismissed. Plaintiffs said that citing the Tennessee Consumer Protection Act was inadvertent and that they meant to plead common-law unfair competition, but they had not filed a motion to amend. They also did not answer Suno’s arguments for dismissing the statutory claim, thereby waiving opposition to dismissal. The court did not decide whether a later amendment should be permitted.
Key Takeaways
- An AI-copyright plaintiff need not always identify a specific infringing output at the pleading stage if detailed factual allegations otherwise make substantial similarity plausible.
- Examples showing that a model can reproduce other protected works may help support an inference about outputs involving the plaintiffs’ own works, though proof will still be required after discovery.
- Alleged stream-ripping that bypasses YouTube’s rolling cipher can state a DMCA section 1201(a)(1) claim before the technical record is developed.
- A party cannot replace an expressly pleaded statutory theory with a different common-law theory merely in its opposition brief.
Why It Matters
The order gives independent artists room to investigate how a generative music model was trained and what it has produced without first locating every allegedly infringing output from outside the company. That is significant because model behavior, historical outputs, and training data may be difficult for copyright owners to observe before discovery.
The DMCA ruling also separates the method of acquiring training material from the ultimate fair-use or infringement questions. Even if later proceedings debate whether training copies are fair use, bypassing an effective access control can present a distinct statutory theory. The ruling is only a pleading-stage decision: it accepts well-pleaded allegations as true and does not establish that Suno infringed copyright or circumvented YouTube’s system.
Your browser cannot display this PDF inline.
Download the full opinion (PDF)