Music publishers amend Anthropic lyrics lawsuit amid author settlement

Generative AI copyright litigation has historically followed a pattern of initial broad complaints followed by either dismissal motions or protracted discovery, with few reaching definitive adjudication on fair use merits. Prior to 2026, most lawsuits aggregated various media types into single infringement theories. The emergence of sector-specific settlements, such as the author-AI agreement, represents a shift toward asset-class segmentation in legal strategy. Historically, music copyright enforcement has operated independently of text licensing due to distinct statutory frameworks and collective management organizations. In AI litigation, this separation is re-emerging as plaintiffs tailor arguments to the specific economic substitution harms unique to their medium. Precedent suggests that when one creative sector settles, adjacent sectors often amend pleadings to differentiate their claims rather than adopting the settled terms. This pattern indicates that 'AI copyright' is not a monolithic legal category but a collection of distinct rights regimes responding differently to technological ingestion.

On July 24, 2026, two significant legal developments converged in the generative AI copyright landscape, highlighting a potential bifurcation in how courts and litigants treat different categories of training data. On this date, music publishers filed an amended complaint against Anthropic, escalating allegations that the company unlawfully trained its models on copyrighted song lyrics without authorization. According to reporting by Music Business Worldwide, this filing was timed closely with a separate judicial approval of a $1.5 billion settlement resolving book piracy claims between AI firms and a coalition of authors [3]. The juxtaposition of these events suggests a diverging legal trajectory: while text-based copyright disputes may be moving toward standardized financial resolutions, music rights holders appear to be intensifying litigation rather than settling.

The amended lawsuit against Anthropic specifically targets the ingestion of lyrical content. Music publishers allege that Anthropic’s models were trained on protected works, seeking damages for what they characterize as infringement. Unlike the author settlement, which resolved claims regarding large-scale text corpora often sourced from shadow libraries, the music publishers’ action focuses on the distinct nature of musical compositions. The amended complaint reportedly expands upon previous allegations, signaling that rights holders in the music industry are not satisfied with current licensing frameworks or voluntary opt-out mechanisms. As of the filing date, Anthropic has not issued a public statement responding to the specific new allegations contained in the amended complaint.

Simultaneously, the federal court’s approval of the $1.5 billion author-AI settlement marks a significant milestone for text-based training data. This agreement resolves claims that AI companies utilized pirated books to train large language models. The settlement establishes a financial precedent for text ingestion, potentially creating a benchmark for future negotiations involving written works. However, the timing of the music publishers' amended filing—occurring on the exact same day as the settlement approval—underscores that the resolution of text claims has not created a universal peace for all creative industries. Instead, it may have clarified the battlefield, prompting music publishers to distinguish their assets from the settled text regime.

Industry observers note that this divergence could lead to separate licensing regimes. Text data, often viewed as functional or factual in the context of LLM training, has now achieved a form of priced resolution through the author settlement. Music lyrics, conversely, carry different cultural and economic weight, with publishers arguing that lyrical ingestion is a direct substitute for licensed lyric display services rather than merely functional training material. The amended complaint against Anthropic tests whether courts will apply the same liability standards to lyrics as they have implicitly accepted for the settled text claims.

The lack of public comment from Anthropic regarding the amended lyrics lawsuit stands in contrast to the high-profile nature of the author settlement. While the author deal provides a roadmap for clearing past text-related liabilities, the ongoing music litigation introduces fresh uncertainty. If the amended complaint succeeds in establishing distinct liability for lyrical training, it could invalidate assumptions that the author settlement represents a comprehensive cap on AI copyright exposure. Conversely, if Anthropic successfully defends against the amended claims, it might reinforce the argument that training use cases differ fundamentally across media types.

This dual-track development also impacts the broader market for AI training data. Data providers and aggregators must now navigate a landscape where text datasets may have a clearer, albeit expensive, path to compliance, while music datasets remain legally volatile. The amended lawsuit serves as a notice that rights holders in non-text domains are actively refining their legal theories to avoid being subsumed under text-centric settlements. For Anthropic and other model developers, the immediate challenge is addressing the specific allegations in the amended complaint while managing the reputational and financial implications of the newly settled text baseline.

Music publishers face continued litigation costs and uncertain outcomes as they pursue amended claims against Anthropic, risking unfavorable precedent that could devalue lyrical assets in AI licensing. Anthropic faces potential damages and injunctive relief that could disrupt model training pipelines for musical content. Authors and AI firms have capped text-related exposure at $1.5 billion, providing certainty for text training but leaving music exposure uncapped. Consumers and downstream developers face fragmented compliance requirements where text datasets may be cleared while music datasets remain legally toxic. The magnitude of divergence means future AI products may have asymmetric capabilities: robust text understanding backed by settled licenses versus constrained musical generation hampered by ongoing litigation risk.