The settlement was supposed to end one fight and start another, and the second one has now begun. Anthropic agreed to pay roughly $1.5 billion to resolve a copyright lawsuit brought by authors who said the company trained its AI models on their books without permission. Before the money can be distributed, the authors are objecting in court to claims on the payout from publishers and literary agents, who they say are trying to take a share of damages that belong to the writers themselves. TechCrunch reported the dispute Saturday.
The argument turns on the language of the contracts that sit underneath the case. The authors say their publishing agreements granted the publishers only the right to use the works in specified ways, and that they never gave anyone permission to feed the books to an AI model. If Anthropic’s training infringed anyone’s copyright, the argument runs, the injury was to the authors, and the compensation should follow the injury rather than the commercial chain that happens to sit above them.
Publishers see the matter differently. Their contracts with authors, built up over decades, typically assign the publishers broad rights to the works, including the right to license and to collect income from uses of the books. In their reading, an AI company that copied millions of books copied the publishers’ property as much as the authors’, and a settlement compensating the copying should reflect that the publishers held the rights the models violated.
Literary agents occupy the middle ground. Agents customarily take a percentage of their clients’ income, and if the settlement is income from the authors’ works, the agents argue, their standard commission applies. The authors counter that a litigation recovery is not the same as a royalty check, and that the agents who did not bring the case and did not bear its costs should not skim the proceeds.
The dispute is a preview of a problem that will follow every major AI copyright settlement. OpenAI and Meta face class actions over the same kind of training data, and any resolutions they reach will raise the same question: when an AI company pays to make amends for copying the written word, who exactly was wronged, the writers who created the books or the companies that own the rights to them?
The answer matters beyond the courtroom. If publishers succeed in claiming shares of AI settlements, future deals will be structured differently, with the companies that control large libraries negotiating on behalf of the authors they represent. If the authors win, the message to writers is that AI compensation belongs to them personally, and the publishers’ role in the digital economy weakens further.
The authors’ position reflects a broader frustration that has been building in the literary world. Writers have watched their income from traditional publishing shrink for two decades, and many believe the contracts they signed in better times were never meant to cover the uses that AI companies made of their work. For them, the Anthropic settlement is one of the few moments of bargaining power the digital era has produced, and they do not want it diluted before the money arrives.
The timing adds a corporate dimension to the legal argument. Anthropic is preparing to go public, and the settlement was widely seen as a piece of pre-IPO housekeeping, a way to clear a legal cloud before presenting the company to investors. The distribution fight does not threaten the deal itself, but it complicates the narrative of a clean resolution, and it ensures the case will stay in the news while Anthropic tries to tell its listing story.
For publishers, the stakes are existential in a quieter way. The economics of the industry have shifted power toward the holders of large backlists, and AI licensing has looked like a new revenue stream for companies whose traditional lines are under pressure. If courts rule that the right to sue over AI training belongs to authors rather than publishers, that stream shrinks, and the companies that have spent years consolidating rights may find those rights worth less than they paid.
The fight will play out in court filings and settlement administration, a procedural arena that rarely makes headlines. But the question underneath it is one the whole industry has been avoiding: in the age of AI, who owns the value of a written work, the person who wrote it or the company that published it? The authors pushing back against their own publishers have forced that question into the open, and the answer will shape how every future AI company pays for the words it trains on.
For now, the authors have filed their objections, the publishers and agents have filed their claims, and the court will decide how the $1.5 billion is divided. Anthropic, having paid to put the copyright fight behind it, finds the fight continuing without it, between the people who created the books and the people who sold them. The company that trained on the words has become a bystander to the argument over who owns them.
