The New York Times and The Daily News have leveled serious accusations against OpenAI, alleging that the company deliberately concealed evidence related to its use of copyrighted materials in training its generative AI models. The newspapers claim that OpenAI's assertions about its inability to search through user interactions and training data are untrue. For two years, the publications have been embroiled in legal disputes with OpenAI, arguing that the company violated copyright laws by utilizing their content without permission, which is allegedly reproduced verbatim in ChatGPT's responses.
OpenAI has consistently maintained that it cannot search its training dataset for specific information and has expressed concerns about user privacy, stating that extracting and anonymizing chat logs would be technically challenging. The publishers have sought access to data that would confirm whether their copyrighted works are included in OpenAI's training datasets and if ChatGPT generates responses that replicate their content.
During a court deposition in April, however, Vinnie Monaco, a data protection engineer at OpenAI, reportedly disclosed that the company did conduct searches within its training dataset to identify copyrighted journalistic works. He mentioned that prior to the lawsuit filed by The New York Times, OpenAI had already compiled a database of 78 million anonymized ChatGPT conversations to assess potential copyright infringements. Following the lawsuit, OpenAI allegedly implemented the Bloom filter within its Project Giraffe toolkit to monitor repeated content in outputs.
Initially, the plaintiffs requested 120 million chat logs, but OpenAI agreed to reduce this to 20 million. When OpenAI provided this data last December, the court found it so heavily redacted that it was deemed "unusable." The plaintiffs further assert that OpenAI deleted several billion ChatGPT responses after the lawsuit was filed, which they argue violated a court order to preserve data, effectively complicating the process of transferring already collected information.
"If OpenAI genuinely believed that copying our clients' journalistic materials was fair and legal, it would not hide the truth about doing so," stated the plaintiffs' attorney. They are requesting the judge to impose penalties on OpenAI for evidence concealment and interference with the discovery process, including barring the use of the 20 million chat logs as evidence, recognizing significant duplications of their materials, and covering legal costs incurred in seeking this evidence.
An OpenAI representative has rejected the allegations, accusing the plaintiffs of attempting to access private user conversations as their legal position weakens. "As the Times' position deteriorates and they are forced to withdraw their claims against us, they continue their attempts to invade the privacy of unrelated individuals, including through these baseless accusations. We will continue to uphold our users' privacy and long-established principles of fair use," the spokesperson emphasized.
The ongoing legal battle highlights significant tension in the AI industry regarding copyright issues and the ethical use of data, raising concerns for competitors who must navigate similar legal landscapes. This case could set a precedent, influencing how AI companies manage proprietary content in their training processes.
Informational material. 18+.
