Musicians sue Google over Lyria AI trained on 44M YouTube clips
In brief
- Musicians sued Google March 6, 2026, claiming Lyria 3 trained on 44 million YouTube clips without consent or payment.
- Google's defense relies on YouTube's terms of service, which grant royalty-free content processing rights.
- The lawsuit could force tech firms to renegotiate AI training practices and reshape user-generated content leverage.
The Complaint
The plaintiffs claim straightforward copyright infringement: Google took their creative work, fed it into a machine learning model, and never asked or paid. Lyria 3, built by Google DeepMind and integrated into Google's Gemini AI chatbot, can produce full tracks complete with vocals and lyrics, lasting up to three minutes. The model represents a significant leap in AI music generation—and a direct threat to independent creators whose work trained it.
Google's defense, filed June 10 via motion to dismiss, rests on a technicality. YouTube's terms of service grant Google a worldwide, non-exclusive, royalty-free license to use, reproduce, and process uploaded content. The company argues that when musicians uploaded their tracks, they agreed to this. The critical question is whether "process" stretches far enough to encompass training a generative AI model that can produce competing music.
The Contradiction at the Heart
Google actively enforces copyright protections on YouTube through its Content ID system, which scans uploads to detect unauthorized use of copyrighted material. That enforcement is selective. The company blocks users from copying one another's work—but claims the right to feed millions of copyrighted songs into its own AI without a single license negotiation. This asymmetry is what the plaintiffs are challenging.
YouTube is the world's largest repository of music, and Google's AI ambitions depend heavily on access to that data. If the company loses the right to train on YouTube uploads without explicit consent, it would need to either negotiate licensing deals at massive scale or find alternative data sources. Neither option is appealing. The scale of licensing required would be enormous. Alternative datasets exist—but they're smaller, less diverse, and less free.
This case will likely reshape how tech firms approach AI training. If courts side with the musicians, the precedent could force renegotiations across the entire industry.
