TechCrunch: NYT says OpenAI hid evidence in ChatGPT copyright trial

In a motion filed in July 2026 and reported by TechCrunch on July 9, the New York Times and the Daily News accused OpenAI of concealing evidence in the ongoing copyright lawsuit over ChatGPT’s training data. The publishers allege that OpenAI falsely claimed it lacked the technical ability to search its training data and conversation logs for copyrighted content — and that internal tools existed which contradicted that claim.

The specific evidence at issue includes a database of approximately 78 million de-identified ChatGPT conversations that OpenAI allegedly used internally to assess copyright infringement levels, and “Project Giraffe,” a set of tools that included a “Bloom” filter designed to detect and record instances where ChatGPT reproduced content verbatim. The publishers argue that OpenAI knew about the extent of infringement but obstructed discovery through data redactions, deletions, and log substitutions.

The case centers on a question with large implications for the news industry: whether AI training on unlicensed journalism constitutes copyright infringement, and if so, what publishers are owed. The existence of internal monitoring tools — if confirmed — would suggest that OpenAI was aware the problem existed, which changes the legal character of the alleged infringement from inadvertent to deliberate.

For journalists and publishers, the significance is practical as well as legal. If the plaintiffs succeed in demonstrating that OpenAI had tools capable of identifying copyright violations and chose not to disclose them, it strengthens the argument that AI companies should compensate publishers for training use. It also raises questions about ongoing training practices at OpenAI and other companies that have made similar representations in litigation.

The case was still in discovery as of the reporting date. No ruling on the sanctions motion had been issued.