Anthropic’s $1.5B Copyright Settlement Has a New Problem

The fight over AI training data is moving from whether creators should be paid to who actually gets the money.

Anthropic’s massive copyright settlement with authors appeared to answer one of the biggest questions hanging over generative AI: what happens when AI companies train models using books obtained illegally?

Now another question is emerging.

Who owns the payout?

Authors have begun raising concerns after receiving notifications showing publishers — and in some cases literary agencies — making claims on payments connected to Anthropic’s $1.5 billion copyright settlement.

The dispute is turning what looked like a landmark resolution for AI copyright into an unexpectedly messy lesson about intellectual-property ownership.

Nearly 500,000 books are involved

The settlement followed a copyright class action against Anthropic.

A judge had previously found that training an AI model on copyrighted books could qualify as fair use, but obtaining those works through piracy was a different matter.

Anthropic ultimately agreed to the $1.5 billion settlement, which received final approval in July 2026.

Under its terms, authors connected to nearly 500,000 titles can receive about $3,000 for each affected work. For traditionally published books that remain in print, the payment is generally split between the author and publisher.

That sounds straightforward.

Publishing rights rarely are.

Old contracts are colliding with new AI money

Some writers say publishers have submitted claims on books whose rights reverted back to the author years ago.

Others say publishers appear to be claiming 100% of payments where authors believe the money should be split.

Literary agencies have also reportedly appeared as claimants in some cases, creating further confusion over who has a legitimate financial interest in the settlement.

The issue may ultimately be more administrative than malicious.

Publishing contracts can stretch across decades. Rights get transferred, reverted, renegotiated and inherited. Databases may not always reflect what the latest contract says.

AI has suddenly put billions of dollars behind those old records.

And inaccurate paperwork now has a very visible price.

AI copyright is becoming an infrastructure problem

This story points to a bigger issue facing generative AI.

The industry initially treated copyright primarily as a legal question:

Can AI companies train on copyrighted material?

Increasingly, it is also becoming an infrastructure question:

Can anyone reliably track who owns what?

Books are only one example.

Music, images, video, journalism and other creative industries all involve layered ownership structures.

If AI companies increasingly license training data or compensate rights holders, the industry will need systems capable of tracking rights at enormous scale.

The companies solving that problem could become surprisingly important parts of the AI ecosystem.

This settlement could influence future AI deals

The Anthropic case is already significant because of its size.

But the mechanics of distributing $1.5 billion may prove equally important.

Future AI licensing agreements will likely need clearer systems for determining who gets paid, how ownership is verified and what happens when multiple parties claim the same work.

The lesson for AI companies is simple.

Paying creators is only one part of the problem.

Figuring out which creator, publisher or rights holder deserves the payment may be considerably more complicated.