Just days after Anthropic announced a landmark $1.5 billion settlement with authors whose books were used to train its AI, a federal judge has thrown cold water on the whole deal — warning he might not approve it at all.

U.S. District Judge William Alsup spent nearly an hour at a hearing in San Francisco on Monday, July 21, tearing apart the settlement he says could force authors into accepting terms they never agreed to. His blunt assessment: "I have an uneasy feeling about hangers on with all this money on the table."

What Happened

Anthropic — the AI company behind Claude — was sued by thousands of authors who accused the company of training its chatbot on hundreds of thousands of copyrighted books, many scraped from pirate websites. Last week, Anthropic announced it had agreed to pay $1.5 billion to settle the case, which was set to go to trial in December.

On paper, it looked like a massive win for authors. In reality, Judge Alsup isn't so sure. He raised sharp concerns that class action lawyers negotiated a deal behind closed doors that authors would have no real power to reject. "They will create a deal behind closed doors and force it down the throats of authors," the judge said, per Bloomberg Law.

The Numbers Don't Add Up — Yet

Under the current settlement terms, authors and publishers would receive roughly $3,000 per covered book. Lawyers for the class action would take home about $101 million. With roughly 465,000 books potentially covered, Judge Alsup wants hard numbers to make sure Anthropic isn't about to get hit with even more lawsuits "coming out of the woodwork."

The judge also questioned whether authors would receive adequate notice about the settlement and their rights. He scheduled another hearing for September 25 to review the claims process — and made clear he'll only approve the deal if his concerns are addressed.

Why This Matters for the Whole AI Industry

This case is being watched closely far beyond the courtroom. It could set a precedent for how AI companies source their training data — and whether they have to pay creators whose work powers chatbots like Claude, ChatGPT, and Gemini.

In June, Judge Alsup ruled that training AI on copyrighted books bought legally counts as fair use — a win for Anthropic. But he also found the company had acquired millions of books through pirate websites, crossing a legal line. The $1.5 billion settlement was meant to cover both the illegally obtained books and the wider class of affected authors.

Now the publishing industry is pushing back hard. Maria Pallante, CEO of the Association of American Publishers, called the judge's attitude "troubling" and said he "demonstrated a lack of understanding of how the publishing industry works." In other words: the fight may not be over.

What Comes Next

The next court date is September 25. Between now and then, both sides will have to address Judge Alsup's concerns — or risk the largest copyright settlement in history falling apart completely. If that happens, the case goes to trial in December, where a jury could decide an even bigger price tag.

For the hundreds of thousands of authors whose books were used to train AI without their permission, the wait continues. The money is still in limbo, and for now, justice is on hold.