We have largely come to accept a modern digital bargain: When we interact with AI chatbots, our prompts become fodder for their ongoing training.
These models feast on a constant stream of user data to survive. Now, however, that massive accumulation of chat history is exactly what has landed OpenAI in hot water.
The New York Times and more than a dozen other publishers asked a federal judge on Thursday to sanction OpenAI over how the company handled evidence in their copyright lawsuit.
The original case asked whether OpenAI trained ChatGPT on stolen journalism. This motion asks something narrower and more damaging: whether OpenAI lied about what it could already prove.
OpenAI accused of obstruction
The filing, submitted in the Southern District of New York, accuses OpenAI of a “deliberate and systemic effort to obstruct discovery,” according to a Bloomberg Law report.
For two years, OpenAI told the court that searching ChatGPT training data and logs for copyrighted material was not technically feasible, a Reuters report noted. Plaintiffs say that was false.
The claim rests on a February deposition of Vinnie Monaco, OpenAI’s privacy engineering lead, a TechCrunch report confirmed.
Monaco reportedly testified that OpenAI had already searched its training corpus and built a database of roughly 78 million de-identified ChatGPT conversations before the Times filed suit in 2023. He also said it had developed an internal tool called Project Giraffe to detect reproduced text.
That timeline undercuts OpenAI’s central excuse, since the tools existed before the lawsuit, not because of it.
OpenAI court case: disputed evidence and deleted logs
Plaintiffs originally sought a sample of 120 million chat logs. OpenAI negotiated that down to 20 million, then submitted a version in December so heavily redacted that the court called it unusable, according to TechCrunch.
The newspapers also allege that OpenAI deleted billions of ChatGPT conversations after a court preservation order took effect.
More OpenAI:
- OpenAI’s $1 trillion ambition could delay its IPO
- OpenAI just built a chip to cut Nvidia out of one job
- OpenAI makes IPO decision amid Anthropic, SpaceX fervor
The remedy plaintiffs want is where the real leverage lies. They are asking the judge to bar OpenAI from using its own reduced log sample as evidence and to simply rule as fact that the logs would show OpenAI reproduced their journalism, Reuters noted.
Ian Crosby, the Times’ lead attorney at Susman Godfrey, said OpenAI concealed what it knew for more than two years, and that resolving discovery this way would settle the case’s most contested technical question without a trial.
OpenAI’s defense and the financial stakes
OpenAI rejects that framing entirely. Spokesperson Drew Pusateri said in a statement that The Times is using privacy claims to compensate for a weakening case, and that the company will keep defending user privacy and fair use.
Notably, the sanctions motion does not target Microsoft, OpenAI’s co-defendant and largest financial backer, according to The Times’ own report on its filing.
The dollar figures already on the table elsewhere in AI litigation show what is at stake. Anthropic agreed to pay authors $1.5 billion to settle a separate training-data case, the largest AI copyright settlement to date, according to The Associated Press.
Related: Your wallet is being put in danger by OpenAI
That amount is a small fraction of Anthropic’s $965 billion valuation, but it set a price for what a court finding against an AI company can cost.
The Times itself has spent more than $28 million fighting AI companies in court, including $4.2 million in this year’s first quarter alone, a Variety report confirmed.
NYT shares traded near $73 this week with little visible reaction to the filing. Discovery motions rarely move markets the way settlements or verdicts do, even when the allegations are this severe.
The bigger stakes as OpenAI eyes a public listing
What the sanctions fight really tests is how AI companies answer for their conduct during litigation, separate from what they did while building their models.
OpenAI is preparing for a public listing that bankers have discussed at valuations approaching $1 trillion, and its eventual prospectus will need to disclose litigation risk like this to new shareholders.
A finding of discovery misconduct would hand publishers, and every publisher watching this case, leverage that a fair use defense alone cannot buy back.
Related: Venice AI raised $65M to exploit OpenAI’s blind spot
