For two and a half years, the fight between the news industry and OpenAI has been a fight about principle: is training a language model on millions of copyrighted articles fair use, or is it theft with extra steps? This week the argument changed shape. In a filing lodged on Thursday in a Manhattan federal court, The New York Times, the New York Daily News and other outlets asked the judge to sanction OpenAI, alleging that the company has been hiding and destroying the very evidence that would settle the question.
The claim is narrow and serious. The newspapers say OpenAI "chose obstruction" rather than hand over the training datasets and ChatGPT logs that could show how copyrighted news was used, and that a recent deposition of an OpenAI employee contradicts what the company has been telling the court for two years. Steven Lieberman, who represents the Daily News and seven sister papers, put it bluntly: the motion asks the court "to punish OpenAI for hiding and destroying evidence showing how ChatGPT was trained on stolen journalism." Among the remedies sought are attorney fees to cover the cost of chasing evidence the plaintiffs say was improperly withheld.
OpenAI's answer is that the restraint is not obstruction but privacy. It has consistently framed its limits on sharing ChatGPT logs as protection for users who never consented to having their conversations trawled through in a copyright case. Spokesperson Drew Pusateri went further, casting the motion as a sign of weakness: "As the Times' case weakens and they've been forced to drop claims against us, they're persisting with their efforts to invade the privacy of people who have nothing to do with this case, including by making these blatantly false allegations."
Both framings can be told with a straight face, which is what makes this the most consequential procedural fight in AI so far. Discovery disputes are usually a sideshow. Here the discovery is the case. If a court cannot inspect what went into a model, then "fair use" becomes an argument conducted entirely in the abstract, with the defendant holding the only copy of the facts. And spoliation findings, when judges make them, tend to be devastating: they can allow a jury to simply assume the destroyed evidence was as bad as the plaintiff says it was.
The backdrop is a copyright landscape that has been shifting fast, and not in one direction. Anthropic settled with book authors for $1.5 billion over training on pirated works, an enormous number in absolute terms and a rounding error against a valuation that has since climbed toward the trillion-dollar mark. That settlement told AI companies something useful: the price of getting this wrong is survivable. The Times is making a different argument to the one the authors made. It is not principally about copying; it is about substitution, the claim that AI companies "free-ride" on the paper's investment in reporting to build products that replace it. Google's AI summaries, which sit above search results and remove the reason to click through, are the proof of concept the newspapers keep pointing at.
The economics are brutal for the plaintiffs' side. The Times has spent more than $28 million on AI litigation, a figure it discloses to financial regulators, and that number includes a separate suit against Perplexity. Very few publishers can spend at that rate, which is one reason so many have taken the other path and signed licensing deals with OpenAI, Google and Meta. The industry is quietly splitting into those who sue and those who sell, and the split is determined less by principle than by balance sheet.
What happens next matters beyond the parties. A sanctions ruling would not decide whether training on news is lawful, but it would decide something arguably more important in the near term: whether the companies building these systems can be compelled to show their work. Dozens of cases are queued behind this one, brought by novelists, artists and record labels, and all of them run into the same wall. You cannot prove what a model learned if nobody outside the lab is allowed to look at what it was fed.