Bartz v. Anthropic explained: the authors settlement and what it means for licensed data

The short answer

How training material was obtained matters, and Bartz v. Anthropic is the clearest public example. In June 2025 the federal judge hearing the case in the Northern District of California ruled on fair use: using lawfully purchased books to train a language model was fair use, and so was scanning purchased print books into digital copies, but downloading and keeping copies from pirate library sites was not excused, and those claims were headed for trial. In September 2025 the parties announced a class settlement of the pirated-copies claims, which then went through the court's approval process.

For anyone licensing data, the lesson is narrow but real. A favorable view of training as fair use did not protect material acquired without permission. Buyers now ask harder questions about where data came from and who had the right to provide it, and records licensed by the company that created them answer those questions directly.

Anthropic is named here only because it was a party to a public court case. Nothing on this page identifies it, or any other party, as a SourceX buyer, client or partner.

What happened in the case?

The dates and holdings above summarize the court's public orders and press coverage; verify each against the case docket before relying on it. For amounts, eligibility rules and deadlines, rely on the official settlement notice and the court's orders rather than press summaries; some deadlines may already have passed. This was one district court ruling on its own facts, and courts in other AI copyright cases can take different approaches; New York Times v. OpenAI is another case to follow.

What does the law actually say?

The case applied fair use, which courts decide case by case. The rules that matter most for companies licensing their own records concern ownership and promises:

• Ownership starts with the author, and rights can be licensed separately. Under 17 U.S.C. § 201, copyright vests initially in the author; for a work made for hire, the employer is considered the author; and any of the exclusive rights can be transferred and owned separately. That is what lets a company license specific uses of material it owns while keeping everything else.
• Employee work usually belongs to the employer. 17 U.S.C. § 101 defines a work made for hire as one prepared by an employee within the scope of employment, or certain commissioned works in listed categories where both parties sign a written agreement.
• The Copyright Office has studied AI training. Its copyright and AI initiative released Part 3 of its report, on generative AI training, as a pre-publication version in May 2025. It addresses where copying in training may implicate copyright, how fair use may apply and how practical licensing is. It is a report, not law.
• Privacy promises are enforceable. FTC staff wrote in January 2024 that companies' promises not to use customer data for undisclosed purposes, such as training models, can be enforced whether they appear in privacy policies, terms of service or marketing (FTC staff guidance, January 2024). That is staff guidance, not a rule.

Why does provenance now carry weight with buyers?

Provenance is the documented answer to three questions: where the data came from, who had the right to provide it, and what the people in it were told. What data provenance is, and why AI buyers pay for it covers the concept in depth.

A license from an operating company builds that record as it goes: an authorized sponsor signs, the scope is written down, de-identification and redaction rules are fixed before preparation starts, and nothing is delivered without an executed agreement and the company's authorization. Compared with scraped or downloaded material, that trail is the point. Buyers' wider expectations are set out in what ethically sourced AI training data means, and the kind of license matters too: training versus retrieval licenses explains the difference.

How does this apply in common partner situations?

What does good disclosure and consent practice look like?

• Keep a written inventory: each system, its date range, who created the content and what is excluded.
• Settle exclusions before anyone prepares data, not after a buyer asks.
• Do not change privacy terms in order to license data without counsel's advice and proper notice.
• Partners never export, upload or describe records; they make the introduction and give basic fit information.
• If you recommend SourceX publicly while earning a referral reward, say so plainly.

Questions to ask your counsel

1. Did the company create these records, or acquire them from someone else, and on what terms?
2. Do client contracts, NDAs or master service agreements limit reuse of documents we hold?
3. What did our privacy policy and customer terms say over the period the records cover?
4. Do our contractor and freelancer agreements assign rights to us?
5. Which folders hold third-party published works that should be excluded?
6. Has any of this data been licensed before, and with what exclusivity?
7. What de-identification do employee and customer details need?

This is general information, not legal, tax or financial advice. Confirm with your own counsel before acting.

What should referral partners take from the case?

The case does not change what a good referral looks like; it sharpens it. Companies whose records were created by their own people, under clean contracts and clear privacy terms, are the strongest candidates. When a founder asks why licensing is different from what they have read about, keep it simple:

The guide on how to explain company data licensing to a founder has more ways to frame it. Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 per referred company, payable only after the buyer pays and SourceX receives its fee; rewards are not guaranteed.

Next step

Run a candidate through the company fit checker. If it looks promising, register as a partner and introduce the company; how it works walks through qualification, inventory, pricing and payment.