Fair Use With No Receipts

The New York Times and allied publishers asked a federal judge on July 9 to sanction OpenAI for discovery misconduct — alleging the company spent two years telling a Manhattan court it could not search ChatGPT output logs for stolen journalism while a deposition from privacy engineering lead Vinnie Monaco revealed it already had, after plaintiffs say OpenAI deleted and substituted millions of logs under a court preservation order — the same day GPT-5.6 Sol went public and OpenAI's IPO roadshow needs investors to believe the training-data fight is manageable.

The filing: The Times, the Daily News, the Center for Investigative Reporting, the Intercept, Ziff Davis, and sister MediaNews papers filed a 52-page sanctions motion Thursday in the Southern District of New York.

They want attorney fees, contempt findings, and jury instructions telling jurors OpenAI destroyed evidence after a judge ordered preservation.

What they allege OpenAI hid: For more than two years, OpenAI told plaintiffs, the public, and the court that searching training datasets and consumer ChatGPT output logs for copies of news articles was infeasible, burdensome, and invasive of user privacy.

In a second deposition in February, Monaco — who leads privacy engineering — said OpenAI had already conducted such searches.

What they allege OpenAI destroyed: After a May 2025 preservation order and a November 2025 production order for 20 million deidentified consumer logs, plaintiffs say OpenAI:

• Substituted roughly two million conversations — about 10% of the court-ordered sample — because the originals were already gone
• Compressed tens of billions of consumer logs on at least 24 occasions, rendering them unsearchable
• Destroyed virtually all API logs and billions of consumer logs tied to user deletions
• Altered conversation IDs and applied over 19 billion redactions, including to URLs pointing at plaintiffs' own articles

OpenAI has signaled deleted data cannot be retrieved.

The quotes that matter:

• Times lead counsel Crosby (Susman Godfrey): OpenAI "lied to The Times, The Daily News Plaintiffs, the public, and the court" and claimed log searches were impossible "while at the same time concealing that it had already done such searches."

• Crosby: "If OpenAI genuinely believed that copying our clients' journalism was fair and legal, it wouldn't have hid the truth about having done it."

• Daily News attorney Steven Lieberman: OpenAI has been "making misrepresentations" for two years about its ability to search training data.

• Lieberman: "This motion asks the court to punish OpenAI for hiding and destroying evidence showing how ChatGPT was trained on stolen journalism."

• The motion itself: OpenAI "chose obstruction" over producing evidence and focusing on fair use.

OpenAI's response: Spokesperson Drew Pusateri said the Times' case is weakening, that plaintiffs are trying to "invade the privacy of people who have nothing to do with this case," and that OpenAI will keep defending user privacy and fair use.

Why this is not the June newspaper pile-on: Nearly 400 outlets filed a coordinated copyright suit in late June. Thursday's motion is a procedural escalation inside the Times-led MDL — discovery warfare, not a new complaint.

The fight is whether OpenAI can ask a jury to bless fair use while the record may show it buried the receipts.

The money: The Times has spent more than $28 million fighting AI companies in court, including a separate Perplexity suit. Variety reported $4.2 million of that burned in the first quarter of 2026 alone.

Plaintiffs want those fees back if the judge agrees OpenAI stonewalled discovery.

The IPO collision: OpenAI confidentially filed an S-1 in June. Anthropic just paid book authors $1.5 billion to settle piracy claims — pocket change against a $965 billion private valuation.

News publishers are testing a different theory: not pirated books, but unfair competition — companies that "seek to free-ride on The Times's massive investment in its journalism by using it to build substitutive products without permission or payment."

A sanctions ruling would not decide fair use. It would tell public-market investors the discovery fight is active, expensive, and getting uglier — on launch day for GPT-5.6.

Convina's view: OpenAI built the world's default information layer, then told a federal court it could not look inside the layer to see whose journalism was copied — until its own privacy engineer said otherwise under oath. Privacy is a real constraint; it is also a convenient shield when the alternative is a searchable log of substitutive outputs. Deleting court-ordered evidence and swapping in substitute conversations is not a fair-use argument — it is a credibility problem heading into a listing where every S-1 risk factor will be cross-examined. If you want investors to price training data as settled law, bring the logs. All of them.