Sony Warner Sue Anthropic: AI Lyrics Copyright Lawsuit

Two record labels are suing Anthropic over song lyrics used to train Claude, and the contested facts are worth noticing even if you never train a model. One side says the material was taken without a licence; the other says training on publicly available data is transformative fair use. Underneath both sits a third claim - that the model will reproduce those lyrics, near enough word for word, when someone asks. Each of those is a statement about what went into a system and what came out of it, and none of them can be settled by looking at the system today. That is why it is in a courtroom rather than a changelog.

If you run agents rather than train them, the same question arrives in a smaller and far more answerable form: what did your agent do, on whose instruction, and can you still show it a year later? That is the record Pragor keeps - who approved what, which agent acted, and what it produced - because the alternative is arguing from memory. This case is about provenance at a scale most teams will never operate at. The habit it argues for is one any team can afford, and it is much cheaper to start before somebody asks.