Anthropic's $1.5B Author Settlement Hits a Judicial Snag

The settlement that was supposed to set the template for resolving AI training-data lawsuits is now stuck in front of a skeptical judge. Ars Technica reports that the federal court overseeing Anthropic's $1.5 billion deal with authors has delayed approval, flagging concerns that plaintiffs' lawyers moved too quickly to lock in roughly $320 million in fees while individual authors are pushing for larger payouts.

The optics are unflattering for everyone involved. Anthropic gets to frame a nine-figure check as the cost of doing business and a clean break from liability. The lead firms get a fee award that dwarfs what most class members will see. The authors, whose books were allegedly ingested without permission to train Claude, are left arguing in court that their own representatives undersold them.

The broader stakes are bigger than this one case. Anthropic's settlement was widely expected to anchor the going rate for unlicensed training data, giving OpenAI, Meta, and others a number to negotiate against in their own pending suits. A delayed or restructured deal — particularly one that raises per-author compensation — would push that benchmark up and complicate the industry's preferred narrative that copyright exposure is manageable and roughly priced in.

It also undercuts the idea that these disputes can be quietly resolved via class action machinery before the law actually clarifies what training on copyrighted text requires. Judges, it turns out, are paying attention to the fee math. The next AI lab hoping to write a check and move on may find that the check has to be larger, and the questions harder to wave away.