ADAM FELDMAN, Legalytics ; The AI Copyright Cases Are Starting to Tell Us Something
"The emerging question is therefore no longer simply whether AI companies can train on copyrighted material. A potentially more useful set of questions is coming into view: What exactly did the defendant copy? How did it obtain the material? What copies did it retain? What can the resulting system reproduce or retrieve? Does the AI product compete with the copyright owner’s market? And what evidence exists that the challenged conduct has actually produced market harm?
The answers do not yet produce a reliable formula for predicting who will win though. Most of the litigation remains unresolved, and motions to dismiss, preliminary injunctions, summary judgments, and settlements answer different legal questions. But the accumulated cases now provide enough variation to identify where the pressure points are developing. And those key points suggest that the next phase of AI copyright litigation may look considerably less like one giant test of AI training than the first wave of cases made it appear.
Two Courts, but an Increasingly Diverse Set of Cases
The litigation is remarkably concentrated geographically. Of the 105 captured U.S. infringement actions, 49 were filed in the Northern District of California and 32 in the Southern District of New York. Together those districts account for 77% of the filings in the dataset.
Consolidation does not explain away the pattern. When related proceedings are collapsed into litigation families, roughly the same share—76%—remains centered in those two courts."
