OpenAI is facing serious accusations from The New York Times and The Daily News, who claim the artificial intelligence giant deliberately concealed evidence related to the use of copyrighted journalism in its ChatGPT outputs. This latest development marks a significant escalation in a two-year-old lawsuit alleging that OpenAI's generative AI models were trained on the news outlets' content and subsequently reproduced that material in user responses.
Throughout the legal proceedings, OpenAI has consistently maintained that it lacked the technical capability to search its extensive training corpus for copyrighted works. The company also argued that retrieving and processing its vast collection of ChatGPT conversations would be unduly burdensome and raise user privacy concerns, requiring complex de-identification procedures. The plaintiffs, however, sought this data to ascertain the presence of their copyrighted journalism within OpenAI's training datasets and to determine how frequently ChatGPT generated responses using or reproducing their content.
A recent court-ordered deposition, reportedly involving OpenAI data privacy engineer Vinnie Monaco, brought to light claims that the company had, in fact, conducted internal searches and evaluations of its training corpus for copyrighted journalism. Furthermore, the deposition allegedly revealed that OpenAI had, even prior to The New York Times filing its lawsuit, accumulated a database of approximately 78 million de-identified ChatGPT conversations. This internal resource was reportedly used to assess the extent of potential infringement. Additionally, a tool named "Project Giraffe," incorporating a "Bloom" filter designed to detect and record content regurgitation in outputs, was allegedly implemented shortly after the lawsuit commenced.
These revelations are particularly critical given previous disputes over data disclosure. The plaintiffs initially requested a sample of 120 million chat logs, which OpenAI negotiated down to 20 million. This sample, submitted last December, was allegedly so heavily redacted that the court deemed it "unusable." The news organisations further accuse OpenAI of deleting billions of ChatGPT outputs after the lawsuit was filed, a direct violation of a court preservation order, and substituting millions of logs within the requested sample. Ian B. Crosby, lead counsel for the plaintiffs, stated, "If OpenAI genuinely believed that copying our clients’ journalism was fair and legal, it wouldn’t have hid the truth about having done it."
In response, The New York Times and The Daily News are now seeking judicial sanctions against OpenAI for allegedly withholding evidence and obstructing the discovery process. They are requesting that the court prevent OpenAI from using the compromised 20 million chat log sample as evidence, accept as fact that ChatGPT logs would have demonstrated significant regurgitation of their content, bar OpenAI from arguing otherwise, and compel the AI firm to cover legal fees incurred in pursuing this evidence. OpenAI spokesperson Drew Pusateri has denied the allegations, accusing The New York Times of attempting to access private user conversations as its case weakens, stating, "We’ll continue defending our users’ privacy and the long-established principles of fair use."