Reply to thread

Message: [QUOTE="Bombastus, post: 91618, member: 2178"] A Massachusetts federal judge ruled on August 20, 2026, that independent artists had plausibly pleaded a Suno derivative-work claim despite identifying no specific infringing output. The ruling kept the claim alive at the dismissal stage rather than deciding that any particular Suno song infringed copyright. Copyright disputes around generative music often mash two different events together. Copying recordings to build or train a model is one alleged act, while producing a generated song that reproduces protected expression is another. Not the same claim. The [B][URL='https://goldmidi.com/community/threads/suno-has-admitted-to-using-youtube-audio-for-ai-training.77034/']way Suno sourced YouTube training audio[/URL][/B] matters to claims about acquisition and training. It does not automatically answer whether a later output unlawfully copies a melody, lyric, recording, or other protected element. [HEADING=2]Training copies and generated songs are separate acts[/HEADING] The major labels made this distinction unusually clear when they sued Suno in 2024. Their original complaint used similar-sounding outputs as evidence that particular recordings had entered Suno’s training data, but expressly said they were not then alleging those outputs themselves infringed unless discovery showed they recaptured parts of the recordings. Their examples were still aggressive. The complaint described 29 outputs resembling “Johnny B. Goode” and 10 resembling “Rock Around the Clock,” including melodic and rhythmic similarities produced with targeted prompts. Yet the legal purpose of those examples was initially evidentiary, aimed at showing what Suno had copied during training. This matters because an output can resemble a source without resolving the separate infringement test. Copyright does not protect every feature of a musical style, genre, production era, or performance manner, so a court has to identify protectable expression and ask whether the challenged work takes enough of it. Training liability can therefore survive even if every public-facing output is eventually found noninfringing. The reverse is possible too. A particular generated track could create an output-side problem even when a broader dispute over how the model was trained remains unresolved. [HEADING=2]Sound recordings have a narrower imitation rule[/HEADING] U.S. law gives sound recordings a special limitation that becomes important with AI soundalikes. Section 114 says the recording owner’s reproduction and derivative-work rights do not extend to a new recording made entirely from independently fixed sounds, even when those new sounds imitate or simulate the copyrighted recording. A synthetic performance can therefore raise different issues depending on what the output actually contains. If it merely recreates a vocal manner, instrumentation, or production feel using new sounds, the sound-recording claim faces a different statutory test from an output that recaptures actual sounds from the master. The underlying composition remains separate. A freshly generated recording can avoid copying the original master while still reproducing protected melody or lyrics closely enough to create a composition claim, which is why [B][URL='https://digitalcommons.law.uga.edu/jipl/vol31/iss2/9/']output similarity in AI-generated music[/URL][/B] cannot be reduced to whether the audio contains a literal sample. No fixed percentage decides substantial similarity. Courts look at protected expression, not a mechanical similarity score, and musical cases can turn on what was copied rather than how many seconds or notes changed. [HEADING=2]Two courts split on how much output detail is enough[/HEADING] The independent artists in Justice v. Suno pushed output infringement further than the labels’ original case. They alleged some Suno outputs were exact or near-exact reproductions of their songs, but they did not identify a particular Suno output tied to a particular plaintiff’s work. Judge F. Dennis Saylor IV still refused to dismiss the derivative-work claim. He relied on allegations about roughly 100 Suno outputs resembling works owned by other copyright holders, claims about how the model was structured, and statements attributed to Suno’s chief executive as enough to support a plausible inference at the pleading stage. A New York judge reached a tougher result against the same artist group in its parallel case against Udio three months earlier. Judge Alvin Hellerstein dismissed the derivative-work count because the artists had not identified specific Udio outputs substantially similar to their own works, although he permitted them to amend. The contrast is more useful than a blanket claim that AI outputs either are or are not infringing. Two federal judges confronted closely related allegations and disagreed about how much specificity plaintiffs needed before discovery, showing that output litigation can turn on pleading detail long before a jury ever compares two songs. Saylor also rejected Suno’s argument that the artists could simply generate suspected infringing songs themselves because the service is public. Whether plaintiffs can make such an output now, he wrote, is different from whether Suno’s system has ever generated one before. Discovery can matter because the legally relevant output may already exist in records the plaintiffs cannot reproduce on demand. [/QUOTE]

Name