Google's Legal Strategy: Deflection Over Disclosure on YouTube-Lyria Training

The Unspoken Confirmation

Google is almost certainly training its Lyria 3 music AI model on YouTube uploads from independent musicians, according to evidence presented in an ongoing lawsuit—yet the company refuses to state this plainly. A motion to dismiss filed by Google in response to litigation by independent creators reveals the company’s legal posture: deflect the factual claim while asserting contractual permission through YouTube’s Terms of Service, thereby avoiding an on-the-record admission that could expose it to further legal challenges.

What the Lawsuit Alleges

According to The Verge AI, a group of independent musicians has sued Google, contending that the company illegally used songs uploaded to YouTube to train Lyria 3. Google’s motion to dismiss takes a two-part defensive stance: first, it argues that the plaintiffs cannot prove Google trained on their specific works; second, even if the company did so, the YouTube Terms of Service grant it the legal right to do so.

The Terms of Service language Google cites permits the company to “reproduce, distribute, [and] prepare derivative works” based on uploads. Google contends this broad license encompasses machine-learning applications—a reading that, if accepted by a court, would insulate the company from liability on contractual grounds, irrespective of fair-use doctrine or copyright protections.

The Pattern of Selective Transparency

Google’s strategic silence on Lyria contrasts sharply with its candor about other AI models. YouTube CEO Neal Mohan told Bloomberg in April 2024 that “some portion” of YouTube videos may be used internally to train models like Gemini. Later that year, a company blog post on creator tools confirmed Google uses content uploaded to YouTube “to improve the product experience for creators and viewers across YouTube and Google, including through machine learning and AI applications.”

The company even confirmed directly to CNBC that it was using YouTube uploads to train both Gemini and Veo. But when asked directly whether YouTube videos were used to train Lyria, Google declined to comment. According to The Verge AI, this refusal is a calculated legal move: the company has little to gain by going on record during pending litigation, and maintaining plausible deniability is a strategic choice designed to minimize exposure.

Why This Matters

This case illuminates a widening gap between what companies do and what they acknowledge doing. For independent musicians and content creators, the lawsuit raises a critical question: does uploading to YouTube constitute informed consent to use for AI training, or does the breadth of YouTube’s Terms of Service exceed what creators reasonably expect or intend?

For platform policy more broadly, Google’s approach—admitting YouTube content trains some models while refusing to confirm others—suggests that corporate transparency around AI training data remains conditional on litigation risk, not principle. If Google’s motion to dismiss succeeds, the Terms of Service argument could become a template for other platforms facing similar claims, potentially shifting the burden of opting out of AI training entirely onto creators rather than platforms.

For music-industry stakeholders and regulators, the case underscores a missed opportunity: if YouTube’s Terms of Service are deemed sufficient to authorize AI training at scale, then the path forward is not litigation but legislative clarity around what constitutes fair use and what requires explicit consent for derivative works in the generative-AI era.