What do the AI copyright lawsuits mean for a company licensing its own records?
Short answer. The US cases so far turn on two points: whether copying works to train a model is fair use, and whether the copies were obtained lawfully. Courts have answered differently depending on the facts, which leaves AI labs with legal risk for material they did not license. A company that owns its records and licenses them in writing gives a lab a clean source, so the cases tend to make well documented company data more useful to labs, not less.
General information, not legal advice.
The questions courts are deciding
Training a model means making copies of the training material. Copyright law gives the owner of a work the right to control copies, so the first question in each case is whether the copying needed a license. The usual defense is fair use, which US courts weigh on four factors: the purpose of the use, the nature of the work, how much was taken, and the effect on the market for the work.
Two facts keep coming up. One is whether the model competes with the work it learned from. The other is where the copies came from, because a court can treat a lawfully bought copy differently from one taken from a pirate site. Neither question has a final answer from the Supreme Court, and different judges have weighed the same factors in different ways.
Thomson Reuters v. Ross Intelligence
Thomson Reuters, which publishes the Westlaw legal research service, sued Ross Intelligence for using Westlaw headnotes to train a legal search tool that would compete with Westlaw. The trial judge ruled in 2025 that the headnotes were protected and that the use was not fair use, and sent the fair use question to the Court of Appeals for the Third Circuit.
In September 2026 the Third Circuit agreed with the trial court. Reporting on the opinion says the panel treated it as an ordinary copyright case: the tool was built to serve the same market as the source, and the court found the use added little that was new. The court also noted that the Ross tool was not generative AI, which limits how far the ruling reaches. Ross has said it will ask the Supreme Court to review the decision.
The book cases in California
In June 2025 a federal judge in San Francisco ruled, in a case brought by authors over books used to train a chatbot, that training on lawfully acquired books was fair use, but that keeping millions of pirated copies in a permanent library was a separate question for a jury. The case then settled as a class action, and the settlement received final approval in July 2026. Another judge in the same court ruled for a different AI developer on fair use in a similar authors case, while saying the outcome rested on the record the authors had built.
The lesson people take from these cases is about sourcing. Copies bought or licensed in the open were treated more kindly than copies pulled from shadow libraries. That is a direct reason for labs to want a written license and a clear account of where each record came from.
What the Copyright Office said
In May 2025 the US Copyright Office released a pre-publication version of Part 3 of its report on copyright and artificial intelligence, covering training. The report described voluntary licensing deals already being made across sectors and recommended letting that market develop, with government action reserved for clear market failure. It also treated harm to a licensing market as relevant to fair use.
The report is guidance, not a ruling, and courts are not bound by it. For a company with records to offer, the point is that the federal office responsible for copyright sees training licenses as a real market that the law should leave room for.
What it means for a company licensing its own records
Most company operating records are not books or news articles. They are tickets, emails, project files and procedures the company wrote for its own work, and the company usually owns the copyright in what its staff produced. A license from that owner removes the fair use question for the lab, because permission replaces the defense.
The cases also show what a lab will check. It will ask who wrote each kind of record, whether any of it belongs to clients or vendors, and how the files were collected. Material the company does not own, such as client deliverables under a contract, purchased reference libraries or articles saved into a shared drive, is either cleared with its owner or left out. Our page on whether selling company data is legal covers the contract and privacy side of the same review. None of this is legal advice; counsel reads the actual records and agreements.
Follow-up questions
Do these lawsuits make it risky for us to license our records?
The suits are about material used without permission. A written license from the owner is the opposite situation. The risk for a seller sits in records it does not own, which is why those are sorted out before anything is offered.
Does fair use mean labs can take our data without paying?
Company records sit inside the company systems, not on the open web, so a lab cannot reach them without the company. Fair use is a defense for copies already made, and courts have split on it, so labs prefer licensed sources.
Who owns the copyright in emails our staff wrote?
Work made by employees in the course of their jobs is generally owned by the employer under US law. Emails from outside senders, contractor work and client files can differ, so counsel looks at each source.
Didn't find your question? Ask us directly.