Sony and Universal hit Suno's freshly licensed v6 with a $9B "poisoned tree" lawsuit, UMG separately sued DistroKid for enabling an AI slop pipeline draining royalty pools, and Alibaba dropped Qwen Audio 3.1 — five models that generate voice, SFX, and background beds in a single pass with API prices slashed up to 95%.
Sony and Universal Sue Suno Again — Call v6 "Fruit of the Same Poisoned Tree"
Nine days after Suno launched v6, its first model trained on music licensed from Warner, BMG, and Believe, Sony Music and Universal Music Group filed a second lawsuit in Boston federal court on September 18. The 45-page complaint names 60,202 sound recordings and carries a theoretical ceiling north of $9 billion. The core argument: Suno trained v6 on the outputs of its earlier, allegedly infringing models, so the new model launders the infringement rather than fixing it. "V6 is not a fresh start; it is the fruit of the same poisoned tree."
This matters because it attacks the cleanup playbook every AI music company is now running — retire the old model, sign licensing deals, retrain from scratch on "clean" data. If the court agrees that outputs from an infringing model taint everything downstream, that strategy collapses. Suno says v6 was trained on licensed catalogs and user preference signals, not audio uploads. The fact discovery deadline is September 30, with a merits decision unlikely before 2027.
UMG Sues DistroKid Over AI "Slop Pipeline" — Seeks $150M
Universal Music Group filed suit against DistroKid in Delaware federal court on September 15, calling the distributor an "AI-slop pipeline" that drains the royalty pool by flooding streaming platforms with undisclosed AI-generated tracks. The 52-page complaint cites specific accounts: one called Lofi Chill uploaded 4,562 tracks in a single year, while technical analysis found that over 97% of tracks from Chill Flow Radio were raw Suno outputs. UMG names 1,000 tracks and seeks up to $150 million in damages.
The lawsuit isn't arguing that AI-generated music should be banned from distribution — it's about disclosure. UMG's complaint targets tracks presented as human-made artist releases when they're not. DistroKid distributes roughly a third of all new music uploaded to streaming platforms. If you're using AI in your production chain and distributing through aggregators, the metadata disclosure question just became urgent.
Alibaba Ships Qwen Audio 3.1 — Five Models, One Audio Stack, Prices Slashed Up to 95%
Alibaba's Qwen team launched Qwen-Audio-3.1 on September 22, a five-model family covering the full audio pipeline: upgraded ASR (with automatic filler-word removal), upgraded TTS (cross-lingual voice transfer with emotion and speed control), upgraded Realtime (full-duplex conversation with empathetic responses), plus two new models. TTS-Next uses a unified LM-plus-diffusion architecture to generate voice, sound effects, and background beds in a single pass — useful for audiobooks, podcasts, games, and ads. ASR-Next handles multi-speaker diarization with timestamps and can caption ambient sounds, detect emotions, and reason about audio content.
The pricing is aggressive: TTS drops ~70%, Realtime ~85%, and ASR up to 95%. If you're building audio-heavy applications and your current provider is ElevenLabs or OpenAI, these prices force a benchmark. The TTS-Next model is the most interesting piece — generating voice and environmental audio together in one pass removes a manual layering step that currently takes real production time.
San Jose Sharks Drop AI-Generated Goal Song After One Season
The San Jose Sharks retired "Power Bite" by BVRNOUT on September 21 after fans spent the entire 2025–26 season calling it AI slop. Online sleuths had identified telltale artifacts suggesting the track was generated with Suno. The team announced a new goal song coming October 1, with defenseman Sam Dickinson theatrically killing the old track in a video. The song played its final two games during preseason on September 22 and 24.
It's a small story with a big signal: audiences are now actively hostile to undisclosed AI-generated music in commercial contexts. Whether or not Power Bite was actually AI-made — neither the Sharks nor BVRNOUT confirmed it — the suspicion alone was enough to force a change. If you're delivering music for brands, sports, or media, the provenance question now carries real reputational risk.
StepFun Releases StepAudio 3 Music — Plans the Song Before It Plays
Chinese AI lab StepFun shipped five StepAudio 3 models on September 15, with the music model standing out for its approach: it writes arrangements in ABC notation first — deciding harmony, structure, and progression — then renders stereo audio up to 5 minutes 30 seconds. This plan-then-render pipeline produces songs with more coherent long-form structure than typical diffusion-based generators. The model supports text prompts, lyrics, vocal input, reference audio, and modes for instrumentals, vocal-to-song, and song covers.
The API (stepaudio-3-music-preview) is free during a limited preview period, with a Gradio-based web studio for quick testing. If you've been frustrated by AI music tools that lose the thread after 90 seconds, the structured-planning approach is worth trying while the preview lasts.
ElevenLabs Ships Music v2.5 — Better Instruments, Full Ownership on Free Tier
ElevenLabs released Music v2.5 on September 11 as the new default in ElevenMusic. The model delivers richer instrument separation and more natural-sounding performances, with blind tests across nearly 48,000 pairs showing preference over v2 across R&B, hip-hop, rock, and orchestral genres. Free users get five lossless downloads per day with full ownership; Pro gets 400 per month. Music v2 stays available alongside it.
The release landed days after ElevenLabs announced a multi-year licensing deal with Universal Music Group, though v2.5 itself was developed separately from that partnership. The ownership-on-all-tiers approach is a competitive move against Suno's more restrictive free tier. If you're evaluating AI music generators for production use, the lossless quality and clear commercial rights make this worth a head-to-head test.
StepFun's Music Studio is a free Gradio web app running the StepAudio 3 Music model. What sets it apart from the current crop of AI music generators is the plan-first architecture: the model writes out an arrangement in ABC notation before rendering, which gives it actual structural awareness across a full 5:30 track. You can feed it text prompts, lyrics, reference audio, or vocal recordings, and switch between instrumental, vocal-to-song, and cover modes.
The output quality sits below Suno v6 and ElevenLabs v2.5 on polish, but the structural coherence over long tracks is genuinely better than anything else at this price point — which is currently free. The API preview won't last forever, so this is a "try it now" situation. Useful for anyone prototyping long-form compositions, game soundtracks, or background music where a full song arc matters more than studio-grade fidelity.
Three developments this month are converging on a single question: how do you prove a track's provenance? Apple Music's upcoming AI Transparency Tags will require distributors to flag tracks where generative AI contributed materially to composition, performance, or production. UMG's lawsuit against DistroKid is fundamentally about disclosure failures, not AI itself. And the Sharks goal song fiasco showed that even the suspicion of AI origin carries commercial consequences.
This is heading toward a world where every track in the distribution pipeline needs verifiable metadata about its creation process — not just "made with AI" or not, but what kind of AI, how much, and with what training data. The tooling for this doesn't really exist yet. Distributors, aggregators, and DAW makers who build robust provenance tracking into their workflows early will have a significant advantage. If you're shipping music through any aggregator, start documenting your production chain now.