AI Copyright Clash: 2 Motions Ask Judge to Kill OpenAI's Fair Use Defense

For AI developers and researchers, the battle over training data just reached a decisive pretrial stage. The motion asks the court to separate the pipeline into acquisition and training phases, arguing that neither is protected fair use—a direct challenge to current data-scraping practices.

On September 4, 2026, a pivotal pretrial contest unfolded in Manhattan federal court, where news organizations led by the Daily News and The New York Times asked Judge Sidney Stein to reject the fair use defense that has anchored OpenAI's and Microsoft's response to claims of mass copyright infringement. The plaintiffs filed a motion for summary judgment seeking a determination that neither company engaged in fair use at any stage of the artificial intelligence pipeline, from acquiring paywalled articles to training large language models on the copied text. The filing drew an immediate framing that is likely to shape the case: the publishers argue that the tech companies are recasting free riding as a lawful privilege, and that the future of journalism and responsible AI both depend on preserving incentives for human creative work.

“

Steven Lieberman, the attorney leading the Daily News' representation, called the material 'scorchingly hot,' saying Microsoft and OpenAI had designated it confidential.

The legal posture is significant because fair use has been the central battleground in the wave of publisher lawsuits against generative AI developers. By asking the judge to decide fair use as a matter of law before trial, the news groups are attempting to strip away the most flexible defense available to OpenAI and Microsoft. A ruling in the plaintiffs' favor would not automatically resolve damages or intentional infringement, but it would convert the remaining trial into a narrower fight over liability, remedies, and what the publishers describe as concealed conduct. The motion specifically isolates the acquisition stage, when systems potentially bypass paywalls, and the training stage, when models ingest millions of articles. The distinction matters because courts have historically been more skeptical of copying that undercuts a direct market for the original work, and paywalled journalism offers one of the clearest examples of market harm.

Alongside the summary judgment motion, the publishers plan to seek the unsealing of heavily redacted filings in the coming weeks. Steven Lieberman, the attorney leading the Daily News' representation, called the material 'scorchingly hot,' saying Microsoft and OpenAI had designated it confidential. The clear implication is that internal admissions by Microsoft executives may undercut the companies' public claims about how their models ingest and reproduce news content. If those admissions become public, they could shape not only this litigation but also settlement dynamics across parallel cases and regulatory inquiries. The plaintiffs have also asked the court to sanction OpenAI over allegedly destroying evidence and concealing its ability to locate news stories in training data and ChatGPT outputs. Judge Stein has not yet ruled on that July motion, and the outcome could have serious consequences for the factual record.