The dust hasn’t settled on Anthropic’s historic $1.5 billion copyright settlement. In fact, it’s getting messier. While the AI company agreed to pay $3,000 for each of the 482,000+ titles illegally used to train Claude, the real battle is now between authors and publishers over who actually gets that money. It’s a chaotic scramble that exposes deep fractures in an industry already struggling with digital rights.
Mary Rasenberger, head of the Authors Guild, points out a systemic issue: many publishers don’t keep accurate records of when rights revert to authors. This administrative gap is costing writers big time. Take April Henry, who found HarperCollins still claiming rights to a book that had long since returned to her control. For textbook authors, the situation is even worse. Some are seeing their publishers offer them just 10 to 15 percent of the payout, despite bearing the brunt of the creative labor. Literary agencies are also jumping into the fray, claiming shares they have no legal standing to collect, according to Writer Beware.
This isn’t just about greedy intermediaries; it’s about broken infrastructure. The court ruled that training on illegally obtained books was unlawful, but deemed using legally purchased ones as fair use. That distinction created this specific pot of money, but the mechanism for distributing it relies on outdated contract data. When disputes arise—and they are arising frequently—a court-appointed arbitrator will step in. But arbitration is slow, expensive, and stressful for individual creators who just want their fair share.
The chaos here reveals a uncomfortable truth about the publishing world’s readiness for the AI era. If major houses can’t track their own rights reverts accurately, how can they possibly negotiate complex licensing deals for future model training? This settlement was supposed to be a clean break, a way to compensate past harms and move forward. Instead, it’s become a stress test for an industry that hasn’t modernized its back office. Until publishers fix their record-keeping, every AI copyright deal risks turning into a legal quagmire for the very people it’s meant to protect.
