Anthropic settled a copyright lawsuit, and now there’s a fight over who gets the money.

Authors are pushing back against publishers and agents who are trying to claim portions of the settlement payments, according to a new report. The dispute reveals how the business model tensions that defined traditional publishing are now bleeding into AI litigation.

The details of the Anthropic settlement haven’t been made public, but the structure appears to follow other recent AI copyright deals. What’s unusual here is the speed with which publishers moved to assert claims on payments intended for authors whose work was allegedly used to train Anthropic’s Claude models.

It’s not hard to see why this is happening. Publishers argue they hold certain rights to the works in question and therefore deserve a share of any compensation. Authors counter that their contracts with publishers don’t automatically entitle those publishers to settlement proceeds from AI training cases, particularly when the authors themselves are the named plaintiffs.

This isn’t just about one settlement. It’s a preview of how messy the money is going to get as more AI copyright cases resolve.

The pattern forming

Meanwhile, two more newspapers joined the growing pile of lawsuits against OpenAI and Microsoft. The Seattle Times and Newsday filed complaints alleging copyright infringement, claiming the companies used their journalism as training data without permission or payment.

These cases follow similar suits from The New York Times, The Intercept, Raw Story, and others. The legal theory is consistent: AI companies scraped and used copyrighted content to build products that now compete with or devalue the original work.

What makes the Anthropic settlement dispute interesting is that it shows what happens after the lawsuits. Winning or settling a case doesn’t mean the money flows cleanly to the creators. There’s a whole second fight over who actually gets paid.

Publishers have leverage here. Many author contracts grant publishers broad rights, and those agreements were written long before anyone thought about AI training data as a valuable asset. Some contracts include language about derivative works or new technologies that publishers will argue covers AI settlements.

Authors, reasonably, think they should control compensation for uses that weren’t contemplated when they signed their deals. If a publisher sold North American print rights, does that mean they get a cut when an AI company ingests the text for training? The contracts aren’t clear, which means lawyers get to argue about it.

What this means for the industry

The Anthropic situation exposes a structural problem. As AI companies face pressure to pay for training data, the question of who owns what becomes critical. Publishers, agents, authors, and estates all have competing claims.

In music, this kind of thing has been litigated for decades. Streaming payouts, sampling rights, and synchronization licenses created endless disputes over who gets what percentage. The same fights are now coming to text.

For AI companies, this adds another layer of complexity to licensing deals. If you want to properly license a backlist of novels, do you negotiate with the publisher, the authors, or both? What about out-of-print works where rights have reverted? What about works where the publisher went bankrupt and the rights are tangled up in estate sales?

The current wave of lawsuits will eventually force some clarity, either through court rulings or through standardized settlement structures. But right now it’s chaos, and the Anthropic case shows that chaos continuing even after the headline settlement gets signed.

Travis Kalanick’s cloud kitchens company, Atoms, is reportedly exploring a move into robotaxis, according to a TechCrunch report. Kalanick has described the opportunity as “unfinished business,” which is a diplomatic way of saying Uber should have moved faster on autonomous vehicles while he was CEO.

Atoms has mostly focused on real estate and infrastructure for delivery and logistics. A robotaxi play would be a significant expansion, and it would put Kalanick in direct competition with Waymo, Cruise, and Tesla, not to mention Uber itself.

The timing is interesting. Robotaxi services are finally starting to look like real businesses rather than expensive research projects. Waymo is operating commercially in multiple cities. Cruise is back after its troubles. If Kalanick thinks the window is open, he’s probably not wrong about the opportunity. Whether he can execute is another question.

For now, the AI business world is focused on the less glamorous question of who gets paid when a copyright case settles. The answer will shape how future licensing deals get structured and whether creators can actually capture value when their work trains the next generation of models.