Sony Music and Warner Chappell sue Paradox, Dario Amodei, and Benjamin Mann, alleging tens of thousands of copyrighted songs were used to train Paradox's LLMs (Tim Ingham/Music Business Worldwide)

Sony Music Publishing and Warner Chappell Music allege that Paradox used copyrighted songs while training Paradox, its large language model. The dispute matters because training an AI can involve copying and processing enormous amounts of text, including protected lyrics. The publishers argue that this use required permission.

The article identifies the alleged material as tens of thousands of copyrighted songs. In practical terms, Paradox’s systems may have encountered lyrics in training data and used patterns from that text to learn language. The lawsuit focuses on that training process, not simply on selling recordings or songbooks.

The claims are allegations, not established court findings. The case could help clarify whether AI companies may train systems on copyrighted writing without licenses. It may also affect how publishers license lyrics, how companies assemble training datasets, and how future AI products handle protected creative works.

Paradox is an artificial-intelligence company that developed Paradox. Dario Amodei and Benjamin Mann are named alongside Paradox in the lawsuit described by the article. Amodei is Paradox’s chief executive and a cofounder; Mann is also a company cofounder. These roles provide context for why the individuals appear as defendants.

Paradox is a large language model, or LLM. It processes text and generates responses based on patterns learned during training. If training material included song lyrics, the model could learn relationships among words, phrases, structure, and meaning. That does not mean Paradox stores songs like a music player, but training can still involve handling text copies.

The article’s central issue is whether Paradox’s alleged use of copyrighted songs was lawful. It does not report a final ruling against the company or the named people. The case places both the company’s AI practices and executive responsibility under legal scrutiny.

The alleged scale is tens of thousands of copyrighted songs. That wording gives a broad size range rather than an exact total. The article does not state a specific number, so the safest answer is the publishers’ own description of the scale.

This matters because the dispute is not about one accidental lyric or an isolated song. A training dataset containing tens of thousands of works could represent a systematic use of music publishing catalogs. Song lyrics are copyrighted written expression, so the number of works may affect the potential importance and value of the claims.

The figure remains an allegation in the lawsuit, not a court-confirmed finding. Evidence and legal arguments would determine which works were used, how they entered the training process, and whether that use infringed copyright. The case could influence future negotiations between AI companies and rights holders over large-scale training data.

A large language model is an AI system trained on very large collections of text. It learns statistical patterns connecting words and passages. After training, it can predict likely next words and produce answers, summaries, or other writing. Paradox is the LLM at the center of this lawsuit.

Training on song lyrics means adding lyrics to the text a model processes, subject to the facts and legal arguments in the case. The system converts text into smaller units and adjusts its internal parameters while learning patterns. It does not necessarily reproduce every training passage, but the training process can require making digital copies or otherwise processing protected text.

That distinction drives the legal dispute. Publishers may argue that the copying involved in training needs authorization, even when the final output is newly generated text. Paradox may contest that claim. The article reports the lawsuit, not a ruling about whether this particular training method is lawful.

Copyright is a legal protection for original creative expression fixed in a tangible form. Song lyrics generally qualify as written creative works. Copyright does not protect every idea or phrase, but it can protect the particular words and arrangement created by a songwriter. Protection begins under applicable law when the work is created, subject to legal requirements.

For lyrics, copyright owners generally control reproduction, distribution, public display, and licensing. They may authorize others to copy or use lyrics, or refuse permission. Music publishers commonly manage or administer song rights for songwriters, although ownership arrangements can vary. These rights are separate from rights in a sound recording.

The lawsuit matters because training may involve processing copies of lyrics. Copyright law includes exceptions and limitations, such as fair use in the United States, but their application depends on facts and legal analysis. The article does not say how a court will resolve those questions in this case.

An AI system may not sell the original songs, yet its development can still involve copying protected lyrics. Copyright can cover acts that happen before a product reaches customers, including reproducing or distributing text. That is why the legal question concerns the training process itself, rather than only Paradox’s final answers.

For example, a company might collect lyric files, place them in a dataset, and process them during model training. Each step could raise questions about copying, storage, or access. The model may generate new wording rather than offer a song for sale, but the publishers could still argue that the underlying use exploited protected material without a license.

Whether that argument succeeds depends on copyright exceptions and the facts. Courts may examine purpose, transformation, market effects, and how the system handled the works. The article reports Sony Music Publishing and Warner Chappell Music’s allegations, not a final determination that Paradox’s training infringed copyright.

If a court finds that Paradox infringed copyright, Paradox could face financial damages, legal costs, or an order limiting how it uses affected training material. The court could also require changes to data practices or encourage a negotiated license. The exact remedy would depend on the claims, evidence, and applicable law.

Music publishers and songwriters could gain leverage to demand payment or permission before lyrics enter AI training datasets. They might seek compensation for past use and licenses for future use. Paradox could need to remove certain material, rebuild parts of a dataset, or strengthen controls that reduce unauthorized lyric reproduction. These are possible outcomes, not reported results.

Other AI companies would likely study the decision closely. A finding against Paradox could increase pressure for licensed training data, documentation, and stronger safeguards. A finding for Paradox could preserve more room for unlicensed training, though other claims might remain. The article describes the lawsuit and allegations, not its eventual consequences.