The New York Times accuses OpenAI of hiding evidence and manipulating the judicial process in the ChatGPT copyright trial

The New York Times and The Daily News have raised the stakes in their two-year copyright infringement lawsuit against OpenAI, formally accusing the company of lying about its ability to search its user conversation logs and training datasets for the presence of protected journalistic content. The original dispute revolved around whether OpenAI trained its generative models on the plaintiffs' articles and whether ChatGPT reproduced that information in its responses to users.

Throughout the litigation, OpenAI maintained two central arguments: that it lacked the technical ability to search its own training corpus, and that producing or reviewing its enormous volume of ChatGPT conversation logs would be excessively costly from a technical standpoint and would raise privacy concerns, since the logs would have to be retrieved, processed and anonymized before they could be analyzed. The plaintiff outlets wanted access to that data precisely to determine whether their journalism was part of the training material and how often the chatbot generated responses based on that content or reproduced it directly.

The turning point came with a court-ordered sworn statement in April, in which Vinnie Monaco, a data privacy engineer at OpenAI, reportedly acknowledged that the company had already conducted internal searches and assessments of its training corpus to locate protected journalistic works. According to that same statement, OpenAI had already assembled, even before the Times filed the lawsuit, a database of some 78 million de-identified ChatGPT conversations, which it used internally to calculate the extent to which it was infringing third-party rights.

A second relevant finding would add to this: shortly after the lawsuit was filed, OpenAI reportedly implemented a filter called "Bloom" within a set of internal tools called "Project Giraffe", designed to detect and record cases of "regurgitation" —that is, verbatim reproduction of content— in the model's responses. For the plaintiffs, the existence of both tools directly contradicts OpenAI's claims of technical inability to perform this kind of analysis.

The procedural backdrop compounds the accusations. The plaintiffs had originally requested a sample of 120 million chat logs, a figure OpenAI negotiated down to 20 million. When the company finally handed over that sample in December, according to the court itself, it contained so many redactions (text censoring) that it was "unusable". The plaintiffs go further and contend that OpenAI deleted billions of ChatGPT-generated outputs after the lawsuit was filed, in direct violation of a court order to preserve evidence, and that it also replaced millions of logs within the court-ordered sample.

Ian B. Crosby, lead attorney for the plaintiffs, summed up the accusation bluntly: "If OpenAI genuinely believed that copying our clients' journalism was legal and constituted fair use, it would not have hidden the truth about having done so". Based on these allegations, the Times and the Daily News are asking the judge to sanction OpenAI for allegedly concealing evidence and obstructing the discovery process. Specifically, they request that OpenAI be barred from using the 20-million-log sample as evidence on the grounds that it is unreliable; that it be accepted as a proven fact that those logs would have shown substantial regurgitation and grounding of the plaintiffs' content; that OpenAI be prohibited from arguing that the logs handed over do not demonstrate that substantial regurgitation; and that the company pay the legal fees incurred in having to pursue this evidence.

OpenAI, for its part, flatly denied the accusations through its spokesperson Drew Pusateri, who accused the Times of trying to access private user conversations at a time when, according to him, the newspaper's case is weakening and it has had to withdraw some of its claims. "As the Times's case weakens and they have been forced to withdraw claims against us, they persist in their efforts to invade the privacy of people who have nothing to do with this case, even making these brazenly false allegations", Pusateri stated, adding that the company will continue to defend its users' privacy and the principles of fair use.

The episode marks a new escalation in one of the most closely watched intellectual property lawsuits within the generative AI industry, with implications that go beyond the specific case: if the court determines that OpenAI did indeed conceal capabilities and evidence it denied having, it could set a precedent on the level of transparency required of AI companies in discovery proceedings, as well as strengthen the position of other media outlets and rights holders pursuing similar lawsuits against the company.