The New York Times and The Daily News allege in an ongoing copyright lawsuit that OpenAI concealed its ability to search its training data and internal chat logs, contradicting claims the company made during discovery.
According to the plaintiffs, an OpenAI engineer's court-ordered deposition revealed the company had already built a database of about 78 million de-identified conversations and deployed tools to detect content reproduction, suggesting OpenAI deliberately made evidence harder to obtain while arguing such searches were infeasible.
What happened
The New York Times and The Daily News claim OpenAI misrepresented its technical capabilities in a copyright lawsuit, alleging that a data privacy engineer's deposition revealed OpenAI had already conducted internal searches of its training corpus and maintained a database of about 78 million de-identified ChatGPT conversations to assess infringement. The plaintiffs say OpenAI also implemented a tool called "Project Giraffe" with a "Bloom" filter to detect and record instances where ChatGPT reproduced content from other sources.
Why it matters
Throughout the lawsuit, OpenAI argued it lacked the ability to search its training data and that doing so would be technically burdensome and raise privacy concerns—the same justification it gave for reducing a requested sample of chat logs from 120 million to 20 million. If OpenAI possessed these capabilities and data all along, it suggests the company made discovery more difficult while withholding evidence it had already collected, undercutting its defense that training on copyrighted journalism was fair use.
What to watch
The plaintiffs are asking the court to deem the 20 million chat log sample unreliable and unusable as evidence, to accept as established fact that the full logs would have shown major regurgitation of their content, to bar OpenAI from arguing the sample demonstrates anything less, and to order OpenAI to pay the plaintiffs' legal fees. OpenAI denies the allegations and accuses the Times of trying to access private user conversations as its case weakens.
Ask the AI about this article →
The New York Times and The Daily News initiated this copyright lawsuit against OpenAI alleging that the company trained its generative AI models on the outlets' journalism without permission and that ChatGPT sometimes reproduces that content verbatim. OpenAI's defense centered on technical and privacy constraints: the company asserted it could not feasibly search its massive training corpus and that retrieving billions of user conversations would be impractical and privacy-invasive. The plaintiffs requested access to 120 million chat logs to verify their claims, but OpenAI negotiated the sample down to 20 million, which the company ultimately submitted with extensive redactions that the court found rendered the data "unusable."
The new allegations, based on an April deposition of OpenAI data privacy engineer Vinnie Monaco, directly undermine the company's stated limitations. If OpenAI had already assembled a database of 78 million de-identified conversations and conducted internal searches specifically to identify infringement on copyrighted works, the company possessed both the technical means and the intent to track the problem—contradicting its position that such analysis was prohibitively burdensome. The revelation that OpenAI had built "Project Giraffe" and its "Bloom" filter to detect regurgitation suggests the company was aware of the issue and monitoring for it internally, yet claimed to courts that providing similar data to the plaintiffs was impractical.
This discrepancy creates substantial litigation risk for OpenAI. The court may view the company's representations as misleading, potentially triggering sanctions (payment of the plaintiffs' legal costs), exclusion of the 20 million chat logs as unreliable, and possible acceptance of the plaintiffs' factual claims by default. OpenAI's denial—asserting that the Times is making "blatantly false allegations" and attempting to invade user privacy—sets up a direct factual dispute that hinges on the credibility of the deposition evidence.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
AI system scaling has pushed interconnect requirements inside data centers from chips and boards up to racks…
Chinese large-model developer Z.ai says it can now support large-scale inference using roughly 100,000 domesti…
Analyst Ming-Chi Kuo says Nvidia has revived the Rubin CPX AI accelerator with a substantially redesigned arch…
Palantir Technologies stock has posted multi-year gains, including an 11x return over 3 years
Apple has escalated its legal battle against OpenAI, claiming in a new court filing that OpenAI is actively de…
Samsung Electronics has locked up as much as 70% of its memory production capacity under long-term supply agre…
