AI Music Digest

Judge F. Dennis Saylor IV denied most of Suno’s motion to dismiss a proposed class action brought by country artist Tony Justice, 5th Wheel Records and My Heartland Publishing, in an order dated August 20 in the US District Court for the District of Massachusetts. Claims for reproduction under Section 106(1), unauthorized derivative works under Section 106(2), and DMCA anti-circumvention under Section 1201(a)(1) all survive. The court granted the motion only as to a Tennessee Consumer Protection Act count, which the plaintiffs said they had cited by mistake and did not defend.

The substantive holding concerns pleading standards. Suno argued the complaint had to identify specific Suno outputs substantially similar to the plaintiffs’ songs, and that it did not. The court disagreed: “there is no categorical requirement that a plaintiff must produce the derivative work at the motion to dismiss stage.” The plaintiffs had instead pleaded roughly 100 examples of Suno outputs replicating or closely resembling songs owned by other rights holders, which the judge treated as supporting a plausible inference about the plaintiffs’ own works — “[w]hether plaintiffs can show such outputs after full discovery is a question for another day.” On the DMCA count, the court adopted its own reasoning from the parallel Universal case.

Why It Matters: Output infringement has been the hardest AI music theory to plead, because plaintiffs rarely hold a matching generated track. This ruling says they do not need one yet — and moves an independent-artist case, not a major label case, into discovery.

According to Music Business Worldwide’s reporting from the docket, Judge Saylor issued two orders on August 18 in the major labels’ case against Suno. He granted Universal and Sony leave to add a claim under DMCA Section 1201(a), alleging Suno circumvented YouTube’s “rolling cipher” to download recordings for training using tools such as yt-dlp. He denied the labels’ motion to add roughly 61,026 recordings to the 560 already at issue, writing that doing so “will have obvious consequences of complexity and delay” and suggesting the labels could bring a separate suit for those works.

The court was explicit that the circumvention question is not settled: “the ultimate determination of whether Suno circumvented a technological measure that effectively controls access to plaintiffs’ sound recordings will require a developed factual record on how the technological measure and circumvention tools work.” Suno had argued the rolling cipher is a copy control rather than an access control, and therefore outside Section 1201(a). The order is also the ruling the judge relied on two days later in the Justice case, which cites it as UMG Recordings, Inc. v. Suno, Inc., No. 24-cv-11611 (D. Mass. Aug. 18, 2026), Dkt. No. 271.

Why It Matters: The labels traded scale for a second theory of liability. Statutory damages on 61,026 recordings would have dwarfed the case; instead they get a circumvention claim that turns on how Suno acquired training data, not on what it generates.

Stability AI announced on August 18 that Stable Audio 3.0 can now be run from inside a digital audio workstation, via a plugin the company is shipping as an early beta. It loads as a macOS AU or VST3 on Apple Silicon and Intel Macs, generates directly onto a track, syncs to session tempo, produces anything from short sections up to six-minute pieces, and keeps multiple takes for comparison. Stability also rebuilt the generation experience on StableAudio.com around iteration — directional text prompts, audio-to-audio variations, and per-track level, pan, mute, solo and effects controls before export. The company has not published pricing for the plugin.

Stability says both are “powered by our commercially-safe models,” and that users “own your outputs, and can distribute outputs freely” — a reference to the licensed training data behind Stable Audio 3.0, which draws on catalog licensed through the company’s Universal and Warner partnerships. That claim is the company’s own; no independent audit of the training corpus has been published.

Why It Matters: Every other route into AI music generation has been a website. Putting a licensed-data model in the instrument slot next to a producer’s synths is a different proposition — and it is the pitch Stability made to the labels that just bought equity in it.

Universal Music Group announced on August 19 a licensing agreement with Hook, an app that lets users remix, mash up and clip existing recordings using AI-driven audio tools. Hook does not generate new songs; it manipulates recordings it has been licensed to use. Under the deal, UMG artists and rights holders decide which songs appear in the app, which creative tools can be applied to them, and where the resulting clips can be shared, and are paid both when fans create with a track and when those creations circulate. Specific rates were not disclosed. The companies say the agreement follows roughly two years of pilot work with UMG labels including Republic, Capitol, Mavin and Virgin Music Group, spanning more than 30 artist campaigns.

Hook CEO Gaurav Sharma framed the product against generative rivals: “The biggest consumer use case for music creation isn’t making original songs that compete with artists.” UMG’s Michael Nash called licensed fan creativity “an important part of the next chapter, combining artist control with responsible technology.” Coverage so far traces to the companies’ own announcement.

Why It Matters: UMG keeps drawing the same line — it will license fans to play with finished recordings while it litigates over models trained to replace them. Hook joins the Spotify remix deal and the Udio platform on the permitted side of that line.