Anthropic settlement payouts: nobody can prove who owns what

The $1.5 billion Anthropic copyright settlement got final approval in July. Paying it out is turning into its own mess. TechCrunch reports today that authors are opening claim-portal emails to find publishers and literary agents have filed competing claims against their payouts — some for books whose rights reverted years ago, some demanding 100% where the settlement allots 50%, some from agents who are not rightsholders at all.

The interesting part is not the greed. It is that an entire industry, with contracts, cannot answer a simple question: who owns this?

What actually happened

The settlement covers roughly half a million works at about $3,000 per work. Per the Authors Guild's breakdown, the default for a traditionally published in-print title is a 50/50 split between the author side and the publisher side. Authors who self-published, hold work-for-hire, or got their rights reverted claim 100%. Everything hinges on the date Anthropic downloaded the pirated library: August 10, 2022. Eligibility keys off copyright registration relative to that date, and rights status as of that date determines who is standing on which side of the split.

That is a clean rule. It is failing on contact with reality because nobody kept the records to apply it. Reversion letters from a decade ago live in someone's inbox. Backlist rights changed hands through three imprint acquisitions. Agency agreements terminated verbally. Where author and publisher cannot agree, the Authors Guild notes the split goes to a special master to adjudicate — a slow, expensive process, per title.

Observers quoted by TechCrunch mostly attribute this to poor recordkeeping rather than bad faith, though the Writer Beware editor flagged the volume of reports as unusual. Both readings land in the same place. The money arrived and the paperwork could not keep up.

Why rights recordkeeping matters for your business

Skip the publishing angle. Ask it about your own company. Somebody trained a model on data you supplied, or scraped your product catalog, or ingested your documentation into a RAG index, and a settlement or licensing pool eventually pays out. Could you prove your claim?

Most small businesses could not, and the reasons are identical to the ones playing out here. The photographer's contract from 2019 is in a former employee's Google Drive. Nobody knows whether the agency that built your site in 2021 assigned copyright or just licensed it. Your ToS says you own user-generated content, but you have never exported the acceptance log.

The controls are unglamorous and cheap:

• Keep a single registry of every content and data asset with its owner, source, license, and effective date.
• Store the signed instrument, not a summary. A row in a spreadsheet is not evidence.
• Record reversions and terminations at the moment they happen, with the date. This is the exact failure mode above.
• When you sign an AI vendor, get the training-data and output-ownership terms in writing, and file them where you can find them in five years.

AI is turning content provenance into a payable asset. That only helps the people who can document it.

Sources: TechCrunch, Authors Guild.