AI
Times alleges OpenAI concealed internal copyright checks in lawsuit over AI training
The New York Times claims OpenAI hid evidence of its own investigations into whether ChatGPT reproduced published journalism in the ongoing copyright dispute.
TechCrunch — AI
The New York Times and The Daily News are escalating their copyright lawsuit against OpenAI, now arguing that the AI company deliberately obscured its ability to detect and track its own use of the outlets' reporting. The accusation marks a significant turn in a two-year-old legal battle over whether OpenAI trained ChatGPT on copyrighted news content without permission.
The hidden capability claim#
At the heart of the dispute sits what OpenAI's data privacy engineer allegedly revealed in an April deposition. According to the plaintiffs' account, OpenAI had conducted internal searches of its training data specifically to identify published journalism—contradicting the company's courtroom assertion that searching its own corpus was impossible. Additionally, the engineer purportedly disclosed that OpenAI had assembled a database containing roughly 78 million anonymized ChatGPT conversations to measure how frequently the model reproduced other people's intellectual property.
The engineers also allegedly deployed a tracking system called the "Bloom" filter, part of a broader initiative labeled "Project Giraffe," designed to flag instances where ChatGPT merely echoed its source material. This system was reportedly activated shortly after the lawsuit commenced.
OpenAI's discovery obstacles#
OpenAI had consistently told the court that providing access to chat logs posed privacy risks and technical challenges that made the task impractical. The company successfully negotiated to reduce the plaintiffs' initial request for 120 million conversation samples down to 20 million. When OpenAI finally delivered that smaller dataset in December, however, the Times and Daily News contend it was riddled with redactions so extensive that the material became worthless to their case.
The plaintiffs also assert that OpenAI deleted billions of chat records following the lawsuit's filing—a move they characterize as violating the court's orders to preserve evidence—and substituted millions of additional logs in the sample.
What the plaintiffs want#
The outlets are now asking the judge to impose sanctions on OpenAI for suppressing evidence and obstructing the discovery process. Their requested remedies include:
