Published 8 hours ago

Anthropic’s $1.5B Copyright Settlement Approved: What It Means for AI Training

A $1.5 billion check is a big number. But for many authors watching this case, it doesn’t feel like a victory lap.

On Monday, Judge Araceli Martinez-Olguin signed off on Anthropic’s copyright settlement, formally closing a case that had been grinding through the courts after findings that Anthropic illegally downloaded and stored millions of copyrighted books. The payout works out to roughly $3,000 per work across an estimated 500,000 titles, split among rights holders.

It’s believed to be the largest copyright settlement in U.S. history. And yet, the creative community’s reaction has been decidedly muted.

3 mins read

5 sections

Key Highlights

Anthropic pays $1.5B after using pirated books from Library Genesis and Pirate Library Mirror
District court ruling treats AI training on copyrighted text as fair use but sets no binding precedent
Google Gemini and OpenAI lawsuits now key tests for how courts will regulate AI training data

Why Authors Aren’t Celebrating

The frustration isn’t really about the money. It’s about what the court actually decided.

Before retiring, the presiding judge ruled that training an AI model on copyrighted text qualifies as fair use. That’s the part the AI industry quietly cheered. The part that got Anthropic into settlement territory was narrower: the company hadn’t just purchased and scanned books (which the court found acceptable). It had also pulled books from pirate sites like Library Genesis and Pirate Library Mirror. That method was ruled illegal on its own terms, independent of the fair use question.

So the settlement resolves the piracy problem. It doesn’t touch the bigger philosophical fight.

A Precedent That Isn’t Really a Precedent

Here’s the structural issue: because Anthropic settled before the case reached an appeals court, the fair use ruling stays at the district court level. It’s one judge’s opinion, not binding law.

Every other judge handling a similar case is free to disagree. And plenty of similar cases are in motion.

Google, Meta, OpenAI, and Midjourney are all facing copyright lawsuits over AI training data. Just last week, a group including Hachette, Cengage, Elsevier, author Scott Turow, and S.C.R.I.B.E. filed a class action against Google over its use of copyrighted works to train Gemini. That case starts fresh, with no obligation to follow Alsup’s reasoning.

What This Means in Practice

For AI companies, the takeaway is fairly clear: how you source training data matters as much as whether you use it. Buying and scanning books survived legal scrutiny. Downloading from pirate mirrors did not. That’s a meaningful operational line, even if the broader fair use question remains unsettled.

For authors and publishers, the settlement offers compensation but no closure. The legal framework governing AI training is still being written, one district court at a time.

The Bigger Picture

The Anthropic case was always going to be a bellwether, not a verdict. It surfaced the right questions—about consent, compensation, and what “fair use” means when the user is a trillion-parameter model—but it didn’t answer them for the industry.

The useful takeaway: watch the Google and OpenAI cases. Those are where the next round of answers, or non-answers, will come from. The fight over AI training and copyright isn’t winding down. It’s just getting started.