Music publisher Round Hill wants $150,000 a song from Anthropic and Suno — and by refusing to settle, it keeps the asking price for unlicensed AI training on the public record

For any business sitting on a content library, a music catalogue, a course, a proprietary dataset, an archive of published research, $150,000 per work has stopped being a figure everyone guesses at and become a filed number somebody has to argue against.

Round Hill is not settling, so unlike disputes that end quietly, its damages math stays where anyone can read it: attached to named songs, named defendants and a date. For a brand or publisher deciding whether to license its own catalogue into a training pipeline, or wondering whether material of its own already went in, that is the difference between negotiating against a number and inventing one. It travels past music, too. Nothing about the claim is specific to songs; it is specific to owning something a model was trained on.

An ask is not an award, and the gap between them is the whole case. $150,000 is the maximum available for a work found willfully infringed, and it is what Round Hill filed for, not what anyone has been ordered to pay. Keep the other separation just as clean: this is a training-data licensing fight, not a citation fight. It changes nothing about whether ChatGPT or Google's AI Overviews quote a client's pages this quarter. The two questions share a root, since both ask whether content owners get paid when their material ends up inside the model doing the answering, but they do not share a calendar.

What the next year decides is whether $150,000 reads as a price or as an opening bid a jury trims. Either way it is the figure that gets quoted in the meantime, including by the person sitting across the table from your client in a licensing conversation.

Recommended actions:

• Ask any client that licenses or distributes proprietary content (courses, datasets, media libraries, published research) one question inside the next 30 days: do they hold any documentation of unauthorized use inside an AI product? Keep the answer in writing, and route anything that looks like a claim to a copyright attorney rather than advising on it yourself.
• Add both filings to the watch list you already keep for the AI-training copyright docket, and log the filing date and defendants in the client-facing content-strategy record, not only in internal notes. The next thing to check for is an exact docket number, which available reporting has not carried yet.
• Keep this separate from AI-search citation economics in every client conversation. If a client hears "$150,000 per song" and asks what it does to their AI Overviews visibility, the answer is nothing: this case is about what goes into the model, not what comes out of it.