Escalating its copyright battle with The New York Times, OpenAI has been ordered to hand over 20 million ChatGPT conversation logs. A federal judge rejected the company’s privacy arguments, ruling that the need for evidence outweighs the risk to user data.
U.S. Magistrate Judge Ona Wang denied OpenAI’s motion for reconsideration on Tuesday, mandating the data transfer within seven days of anonymization. The ruling provides the Times with significant ammunition to rebut claims that it manipulated the AI to generate infringing content.
Despite warnings from OpenAI’s security chief that the move “breaks with common-sense security practices,” the court found internal safeguards sufficient. The company has immediately appealed the decision to a district judge.
Promo
The Ruling: Privacy Shield Pierced
Magistrate Judge Ona Wang issued a formal denial of OpenAI’s motion for reconsideration regarding the discovery dispute. In a detailed nine-page opinion and order, the court explicitly rejected the argument that user privacy concerns should override the evidentiary needs of the plaintiffs.
Far from a simple procedural hurdle, the order represents a significant piercing of the “black box” protection often claimed by tech firms regarding user data.
Citing the “proportionality” standard of Federal Rule of Civil Procedure 26, the judge found that the sheer volume of data did not constitute a valid reason to withhold it.
Addressing the balance between confidentiality and legal discovery, Judge Wang stated that “The Court recognizes that the privacy considerations of OpenAI’s users are sincere. However, such considerations are only one factor in the proportionality analysis, and cannot predominate where there is clear relevance and minimal burden.”
Judicial reasoning relied heavily on the relative scale of the request compared to the company’s extensive data holdings. As detailed in the court’s filing, Judge Wang noted:
“The total universe of retained consumer output logs is in the tens of billions. The 20 million sample here represents less than 0.05% of the total logs that OpenAI has retained in the ordinary course of business.”
“Moreover, the burden of production is minimal at this point; OpenAI has completed (or nearly completed) the extensive process of applying its internal de-identification tool—which OpenAI itself has lauded as significantly more effective at removing both personally identifiable information and private information.”
Magistrate-Judge-Wang-Order OpenAI New York Times 20251202
Compliance requires strict adherence to a timeline. Dismissing the notion that the production creates an “undue burden,” the judge noted that OpenAI has already completed most of the technical work required to sanitize the data.
Establishing a hard deadline, the court ruled that “OpenAI is directed to produce the 20 Million ChatGPT Logs within 7 days of completing the de-identification process.”
Such a decision sets a potential precedent for how AI user data is treated in litigation. By ruling that internal de-identification tools are sufficient to mitigate privacy risks, the court has signaled that AI companies cannot easily use “user privacy” as a blanket shield against copyright discovery.
Strategic Stakes: The ‘Hacking’ Defense
Obtaining these logs is essential for The New York Times to dismantle OpenAI’s primary defense: the accusation of “hacking.” In previous filings, the AI company alleged that the Times used manipulative prompts to force ChatGPT to regurgitate copyrighted articles, creating artificial infringement rather than exposing a systemic flaw.
Central to the initial copyright lawsuit filed in December 2023, this legal maneuver alleged that the Times used manipulative prompts to force ChatGPT to regurgitate copyrighted articles.
By analyzing “real” user queries, the Times hopes to prove that the model spontaneously reproduces copyrighted text without adversarial prompting from researchers. Validating this investigative approach, the court’s order states:
“Thus, the 20 Million ChatGPT Logs are clearly relevant to News Plaintiffs’ output claims to the extent that they contain partial or whole reproductions of News Plaintiffs’ copyrighted works, and to OpenAI’s affirmative defenses to the extent that they contain other user activity—and News Plaintiffs are entitled to discovery on both.”
OpenAI’s legal team had argued that 99.99% of these logs are irrelevant to the copyright claims, a figure the court found unpersuasive without proof. Plaintiffs are also investigating the “Pink Slime” theory, which suggests that AI floods the market with low-quality derivatives of high-quality journalism, diluting the value of original reporting.
Frank Pine, executive editor of MediaNews Group, which is part of the consolidated lawsuit, criticized the company’s resistance to transparency. Highlighting the friction between the parties, Pine remarked that “OpenAI’s leadership was hallucinating when they thought they could get away with withholding evidence about how their business model relies on stealing from hardworking journalists.”
Discovery is now shifting from theoretical arguments about “fair use” to concrete forensic analysis of model behavior. Access to these logs will allow the plaintiffs to see how often regular users—not just Times investigators—are served copyrighted content.
The Privacy Debate & Technical Safeguards
OpenAI’s security leadership has reacted strongly to the order, framing it as a significant overreach into user privacy. Arguing that even anonymized data can be re-identified, the company warns of risks regarding unique personal queries or writing styles.
Dane Stuckey, OpenAI’s Chief Information Security Officer, publicly criticized the demand. Defending the company’s stance, Stuckey stated that “The Times’ demand for the chat logs disregards long-standing privacy protections and breaks with common-sense security practices.”
However, the court turned OpenAI’s own marketing against it. Judge Wang cited the company’s previous claims about the effectiveness of its de-identification tools as a reason to trust the process.
This follows a previous preservation order in July, where the court forced OpenAI to retain all deleted chats to prevent the destruction of evidence.
To further mitigate risks, the logs will be subject to an “Attorneys’ Eyes Only” designation. This legal classification theoretically prevents the data from leaking to the public, the press, or even the plaintiffs’ own business executives.
Emphasizing these judicial safeguards, Judge Wang noted that “There are multiple layers of protection in this case precisely because of the highly sensitive and private nature of much of the discovery.”
Distinguishing this case from Nichols v. Noom, a prior case where privacy concerns limited discovery, the ruling noted that the protections available here justified the disclosure. Judge Wang emphasized that the specific relevance of the logs to the “fair use” defense outweighed the privacy risks.
The Broader Legal War
OpenAI has immediately appealed the Magistrate Judge’s ruling to District Judge Sidney Stein, seeking a stay. Overturning such an order requires meeting a high standard; the company must prove the ruling was “clearly erroneous” or contrary to law.
At the heart of the dispute lies a fundamental disagreement over the economics of AI training. Financial stakes are escalating rapidly, with The New York Times reporting over $7.6 million in legal fees for the first nine months of 2024 alone.
Steven Lieberman, an attorney for The New York Times, reiterated the core economic argument driving the litigation. Framing the case as a matter of theft rather than innovation, Lieberman said in March that “We appreciate the opportunity to present a jury with the facts about how OpenAI and Microsoft are profiting wildly from stealing the original content of newspapers across the country.”
Representing just one front in a multi-district litigation, this discovery battle includes authors, other publishers, and potentially class-action claimants. While some publishers like The Washington Post have opted for content partnerships, the “litigation coalition” is digging in for a long, forensic trial.
The outcome of this specific dispute could force other AI companies, such as Google and Anthropic, to prepare for similar disclosures. If the Times successfully uses these logs to prove systemic regurgitation, it could undermine the “fair use” defense that underpins the entire generative AI industry.