NYT Accuses OpenAI of Hiding Evidence in Major Copyright Trial
The ongoing legal battle between The New York Times and OpenAI has taken a dramatic turn, with allegations that the AI giant intentionally obscured evidence regarding its training datasets. This escalation threatens to shift the focus of the lawsuit from copyright infringement to the integrity of the judicial discovery process.
Allegations of Deceptive Data Practices
The New York Times and The Daily News claim that OpenAI has been untruthful regarding its technical capacity to search both its training corpus and customer chat logs. Throughout the two-year litigation, OpenAI has maintained that searching its massive datasets would be technically burdensome and would compromise user privacy.
However, a recent deposition from OpenAI data privacy engineer Vinnie Monaco has challenged this narrative. Monaco allegedly revealed that OpenAI had already conducted internal searches of its training corpus specifically to identify copyrighted journalistic works. Furthermore, the deposition suggests that OpenAI had amassed a database of approximately 78 million de-identified ChatGPT conversations to internally assess the extent of potential copyright infringement.
Project Giraffe and the "Bloom" Filter
Perhaps the most significant technical revelation involves "Project Giraffe," a set of tools implemented by OpenAI. According to the plaintiffs, shortly after the lawsuit was filed, OpenAI utilized a "Bloom" filter as part of this project to detect and record instances of "regurgitation"—where the model reproduces copyrighted content verbatim in its outputs.
The existence of Project Giraffe contradicts OpenAI's previous stance that identifying specific instances of content reproduction within its logs was a prohibitive task. The plaintiffs argue that these tools prove OpenAI was not only capable of monitoring copyright infringement but was actively building infrastructure to track it.
Disputes Over Data Integrity and Discovery
The conflict has also centered on the quality of the data provided to the court. While the plaintiffs originally requested a sample of 120 million chat logs, OpenAI negotiated the figure down to 20 million. When the sample was finally submitted in December, the court reportedly found it "unusable" due to excessive redactions.
The NYT has gone further, alleging that OpenAI deleted billions of ChatGPT outputs in violation of a court-ordered preservation order and substituted millions of logs in the provided sample. In response, OpenAI spokesperson Drew Pusateri has denied all allegations, accusing the Times of attempting to invade user privacy as their legal case weakens.
Why This Matters for the AI Industry
This case is a bellwether for the future of generative AI development and the legal boundaries of "fair use." If the court finds that OpenAI withheld evidence or manipulated discovery, it could set a precedent for how AI companies are required to disclose their training methodologies and data lineage. For developers and AI founders, the outcome will clarify the level of transparency required when navigating the intersection of massive data ingestion and intellectual property rights.
Key Takeaways
- Internal Monitoring Tools: Allegations suggest OpenAI used "Project Giraffe" and a "Bloom" filter to actively track the regurgitation of copyrighted content.
- Contradictory Testimony: Deposition evidence suggests OpenAI possessed the technical ability to search its training data, contradicting its previous claims of technical burden.
- Discovery Integrity at Stake: The lawsuit now hinges on whether OpenAI violated preservation orders by deleting outputs and providing "unusable" redacted data samples.