News organizations suing OpenAI, led by The New York Times, filed a sanctions motion Thursday accusing the company of misleading a federal court for two years about its ability to search ChatGPT conversation logs. Ars Technica reported that the filing centers on testimony from OpenAI privacy engineer Vincent Monaco, who was ordered back for a second deposition in April after the court found his first appearance inadequate.
According to the plaintiffs’ filing, Monaco testified that OpenAI held two large samples of de-identified logs, one spanning 10 million conversations and another spanning 78 million, months before the case reached its current discovery fight. The plaintiffs allege OpenAI had already searched those samples for New York Times content while building an internal filter meant to block regurgitation of copyrighted text. None of that was disclosed to the court.
The distinction matters because the sanctions fight turns on cost and feasibility claims OpenAI made early in the case. The plaintiffs say OpenAI told the court that broad log searches were technically infeasible and would violate user privacy, then negotiated instead for an eight month process in which plaintiffs could only search a redacted 20 million log sample. That sample was smaller than the 120 million logs plaintiffs originally sought, narrowed, the filing says, by representations about search capability the plaintiffs now call false.
The plaintiffs’ filing describes a second problem with that same sample. OpenAI applied 19 billion automated redactions to it, a volume the court reportedly deemed the resulting data “unusable” before some redactions were later reversed. Separately, plaintiffs allege OpenAI deleted or compressed billions of logs that a court preservation order required it to keep, and that Monaco testified the company weighed compliance and decided against it because doing so “would be hard.”
These are allegations, not court findings, and OpenAI disputes them. A company spokesperson told Ars Technica the sanctions motion is a late attempt to access more user data and framed the Times’ recent decision to drop certain claims in the case as evidence its lawsuit is weakening, not OpenAI’s defense. NYT spokesperson Graham James pushed back on that framing last month, telling Ars the suit was streamlined rather than weakened after plaintiffs added copyright claims against Microsoft. NYT lead counsel Ian Crosby said in a statement that OpenAI “lied” about search feasibility while concealing that the work had already been done.
The plaintiffs are asking the court for consequences well beyond a fine. They want the redacted 20 million log sample excluded, a court finding that the withheld logs contained substantial regurgitation of copyrighted material, and a jury instruction stating that OpenAI deleted billions of logs it had been ordered to preserve. Each request would hand the plaintiffs a shortcut past two years of discovery disputes rather than requiring them to prove infringement from scratch.
If the court accepts that OpenAI’s conduct was willful, the ruling could reshape how judges police discovery obligations across the growing docket of AI copyright suits, not just this one. A spoliation-based jury instruction tends to color how a jury weighs everything else a defendant argues about fair use, regardless of the underlying merits. For any company defending a training dataset in court, this motion is a preview of what it costs when a judge concludes that discovery answers were shaped strategically instead of accurately. OpenAI’s next filings, expected before a ruling on sanctions, will show whether it can rebut the timeline Monaco’s testimony laid out.
Reporting by Ashley Belanger for Ars Technica, published July 9, 2026.
