Google Asserts YouTube License Covers Lyria 3 Training

Independent musicians sued Google, alleging the company trained its music-generation model on copyrighted recordings they uploaded to YouTube, according to the complaint and a press release hosted by plaintiffs' counsel Loewy, as reported by Music Business Worldwide and other outlets. Per reporting, Google filed a motion to dismiss on June 8 in the US District Court for the Northern District of Illinois; the motion, filed by Quinn Emanuel, argues that plaintiffs granted YouTube and its "Affiliates" a "worldwide, non-exclusive, royalty-free, sublicensable and transferable license" that authorizes the uses alleged, a point summarized in Music Business Worldwide and Billboard. The motion also contains the line that the plaintiffs rely on an "unsupported hypothesis" that Google trained on their specific works, as quoted in the filings and reported by The Verge and Music Business Worldwide.

Per DeepMind's public product pages, Lyria 3 is a multimodal music generation system designed to produce up to 3 minutes of high-fidelity audio and accepts text, audio, and images as inputs. CNET and DeepMind documentation state that Lyria 3 and Lyria RealTime are integrated into Google product surfaces including Gemini and Google AI tools, and that the system can generate structured song elements (intros, verses, choruses) at higher fidelity in the Pro tiers.

Industry observers: Litigation over model training data frequently turns on two legal axes: copyright doctrine (including fair use) and contractual licenses in platform terms of service. Reporting on this case emphasizes the latter, with Google pointing to the express license language in YouTube's ToS. Previous AI training lawsuits have often focused on fair use; this case differs in that the defendant is asserting a contract-based defense, as noted in Billboard and Music Business Worldwide.

Editorial analysis: If courts accept contract-language defenses where platforms obtain broad sublicensable rights from uploaders, that could shift the legal landscape away from purely fair-use debates toward contract interpretation for a range of training-data disputes. Reporting highlights that YouTube already pays substantial sums to the music industry, with Music Business Worldwide noting YouTube paid more than $8 billion to the music industry in the twelve months to June 2025 - an economic context the filing cited.

Editorial analysis: Observers should track:

• •how the court interprets the specific ToS language about "Affiliates" and sublicensability
• •whether the plaintiffs can plausibly allege model training used their particular recordings despite broad platform licensing clauses
• •whether other courts treat platform ToS as dispositive in data-training disputes. The plaintiffs listed in reporting include New York singer-songwriter Sam Kogon, Los Angeles composer Magnus Fiennes, and Atlanta producer Mi[redacted in some reports], per Music Business Worldwide and Loewy filings. The Verge reports Google declined to comment when asked specifically about using YouTube content to train Lyria 3

For practitioners: Dataset collectors, platform engineers, and counsel building training pipelines should note the dual legal risks highlighted by reporters: copyright liability and contract limitations embedded in upload terms. Even when training datasets are technically feasible to assemble, public platform agreements and their judicial interpretation can shape permissible uses. Researchers and product teams relying on platform-derived corpora will likely need to monitor evolving case law and platform policies closely.