OpenAI faces serious allegations that it misled courts about its technical capabilities to search its own training datasets and ChatGPT conversation logs for copyrighted material, according to claims filed by the New York Times and Daily News in their ongoing copyright lawsuit. The revelations suggest the company may have deliberately obscured evidence that could prove how much copyrighted journalism it used to train its AI models.

What Evidence Did OpenAI Allegedly Hide?

In an April court-ordered deposition, an OpenAI data privacy engineer named Vinnie Monaco allegedly revealed that the company had already conducted internal searches and evaluations of its training corpus to look for copyrighted journalism works. This directly contradicts OpenAI's earlier arguments to the court that searching its massive training dataset would be technically impossible and would raise privacy concerns.

The allegations go further. According to the Times and Daily News, OpenAI had amassed a database of approximately 78 million de-identified ChatGPT conversations that it was using internally to measure how much it was infringing on others' works. Additionally, the company allegedly implemented a tool called "Bloom" filter as part of "Project Giraffe," which detected and recorded instances where ChatGPT regurgitated copyrighted content in its responses, shortly after the lawsuit was filed.

How Did OpenAI Obstruct the Discovery Process?

The plaintiffs claim OpenAI made it unnecessarily difficult for them to obtain information the company had already collected. Here's what allegedly happened during the legal discovery process:

• Reduced Sample Size: The Times originally requested 120 million chat logs, but OpenAI negotiated to provide only 20 million instead.
• Heavy Redactions: When OpenAI finally submitted the 20 million chat log sample in December, it included so many redactions that a court described the sample as "unusable."
• Deleted Outputs: The plaintiffs claim OpenAI deleted billions of ChatGPT outputs after the lawsuit was filed, allegedly violating a court preservation order.
• Substituted Logs: OpenAI allegedly substituted millions of logs in the requested sample, further compromising the integrity of the evidence.

"If OpenAI genuinely believed that copying our clients' journalism was fair and legal, it wouldn't have hid the truth about having done it," stated Ian B. Crosby, lead counsel for the plaintiffs.

What Are the Plaintiffs Asking the Court to Do?

The New York Times and Daily News are requesting several remedies from the judge to address what they characterize as evidence tampering. They want the court to prevent OpenAI from using the 20 million chat log sample as evidence, claiming it is unreliable. They also want the court to accept as established fact that the full ChatGPT logs would have shown substantial regurgitation and copying of the plaintiffs' content. Additionally, they are asking OpenAI to pay their legal fees for having to pursue this evidence.

How Is OpenAI Responding?

OpenAI has denied the allegations and accused the Times of attempting to invade the privacy of users who have nothing to do with the case. "As the Times' case weakens and they've been forced to drop claims against us, they're persisting with their efforts to invade the privacy of people who have nothing to do with this case, including by making these blatantly false allegations," said Drew Pusateri, an OpenAI spokesperson. The company reiterated its commitment to defending user privacy and what it characterizes as "long-established principles of fair use."

This escalation marks a significant turning point in the two-year legal battle over whether OpenAI violated copyright law by training its generative AI models on the Times' journalism without permission and reproducing that content in user outputs. The case has become increasingly contentious, with both sides accusing the other of bad faith conduct. The court's response to these allegations could fundamentally shape how AI companies handle copyright claims and data transparency going forward.