A federal judge just derailed what copyright lawyers are calling a landmark victory — and the reason exposes a troubling pattern in how AI companies settle claims from the creators they trained on.
On Thursday, US District Judge Araceli Martinez-Olguin declined to approve Anthropic's record $1.5 billion settlement with authors whose books were allegedly pirated to train the company's AI systems. Her concern: the deal may enrich lawyers far more than the writers it supposedly compensates. Ars Technica reported that objecting class members described their potential payouts as a "pittance."
The settlement represents the largest copyright settlement in US history, yet critics argue the structure reveals a fundamental imbalance. Attorney fees could reach into the tens of millions of dollars — a windfall for a legal team that spent years litigating rather than creating. Individual authors, by contrast, might receive sums Anthropic's own training data was arguably worth billions to generate.
"This proposal insults our contribution," one objecting author wrote in a letter Ars reviewed. That sentiment captures the central tension: Anthropic built Claude partly on their work, yet the settlement would see most money flow elsewhere.
The implications extend beyond this single case. If courts bless settlements where AI companies pay enormous headline figures while creators receive pennies, the precedent could incentivize a new playbook. Pile on enough legal costs, bury the math in complexity, and a $1.5 billion settlement becomes effectively cheap. Anthropic gets regulatory cover and cleared training data; authors get fractions of pennies per word.
Judge Martinez-Olguin has signaled she wants to hear directly from objectors before deciding whether this settlement genuinely serves the class — or whether it's a sophisticated wealth transfer from AI companies to the lawyers who sue them. Her questions could reshape how courts value training data going forward. The 15 billion-dollar question is whether anyone besides lawyers wins.