Lawsuit: Google accused of training AI on music from YouTube

Google is trying to get a lawsuit from a group of independent musicians thrown out—but its legal defense is raising fresh questions about how its Lyria 3 AI music model was trained. That’s according to reporting from, among others, The Verge.

Court filings show Google arguing that YouTube’s terms of service grant the platform a broad license over uploaded content. That implies the company may be entitled to use uploads for AI training—without explicitly confirming that this happened for Lyria 3.

The case hits one of generative AI’s biggest flashpoints: can tech companies use user content on their own platforms to train models?

A group of independent artists alleges Google used millions of copyrighted songs from YouTube to train Lyria 3, the company’s latest AI music model. They say they never consented and received no compensation.

Google asked the court to dismiss the case, arguing there’s no evidence the plaintiffs’ specific tracks were included in the training data. The company also leans on YouTube’s terms, which grant the platform a broad license to reproduce, distribute, and create derivative works from uploads.

That legal stance is striking. In essence, Google argues: even if the music was used, it could fall within the granted rights.

The debate is messier because Google has previously acknowledged using YouTube content for AI development.

In 2024, YouTube CEO Neal Mohan told Bloomberg that some YouTube videos are used to train AI systems like Gemini. Google later confirmed uploads help develop Gemini and the video-generation model Veo. An official YouTube blog post also states that uploaded content is used for machine learning and AI applications across Google products.

What Google won’t say outright is whether the same applies to Lyria 3.

According to The Verge, that’s likely deliberate. A formal admission could weaken Google’s position in ongoing copyright cases. For now, the company is leaving the door open—without stepping through it.

Lyria 3 is one of Google’s most advanced AI music models. It can generate full music clips—vocals, instrumentation, and lyrics—from text, images, or video prompts. Since early 2026, it’s been integrated into Gemini.

Because music is tightly protected by copyright, training data provenance is under intense scrutiny. Artists and rights holders want to know if their work is being used without consent to build systems that can then produce competing music.

Google has said Lyria is trained only on music it has rights to via agreements, platform terms, or applicable law. But it has never publicly disclosed the exact datasets.

The Google suit is part of a global wave of litigation over AI training on copyrighted material. Publishers, artists, photographers, programmers, and musicians are demanding transparency around training data—and fair compensation when their work fuels AI.

For Google, the stakes are higher: it owns both the AI models and YouTube, one of the world’s largest repositories of creative content. Plaintiffs argue that gives Google access to vast music libraries competitors can’t easily match.

The outcome could reshape the AI industry. If a court rules that platform terms provide sufficient consent for AI training, tech companies gain a powerful legal precedent.

A different ruling could trigger stricter rules, higher licensing costs, and tougher transparency requirements for training data.

For millions of creators on platforms like YouTube, the core question is ownership: does uploaded content remain solely the creator’s—or can a platform use it to build new, revenue-generating AI products? That question is now squarely before the court in a case centered on Google’s music AI.