News Publishers Claim OpenAI Concealed Training Data in Copyright Disputes

On Thursday, a coalition of 17 news organizations, including The New York Times, New York Daily News, Chicago Tribune, and Ziff Davis (parent of CNET), filed a motion alleging that OpenAI is withholding evidence related to how its AI models are trained, in connection with ongoing copyright litigation.

Ziff Davis initiated its lawsuit against OpenAI in 2025, claiming the company scraped its copyrighted material to train ChatGPT and other large language models. The original suit, launched in 2023 by The New York Times against OpenAI and Microsoft, asserted that the firms built their AI systems using millions of journalist‑written articles without permission; both defendants have denied the allegations.

The motion seeks court‑imposed sanctions against OpenAI (but not Microsoft) for allegedly failing to produce datasets, output logs, and other training‑related evidence, arguing that OpenAI “chose obstruction” by withholding this information. If granted, sanctions could include financial penalties. Steven Lieberman, attorney for the New York Daily News, told the Associated Press that the motion asks the court to penalize OpenAI for hiding and destroying evidence that shows ChatGPT was trained on stolen journalism.

Central to the lawsuits is the question of how generative AI systems such as ChatGPT are trained and where they obtain their data. The Times’ original complaint contends that OpenAI’s tools can reproduce its content verbatim, produce close summaries, and imitate its expressive style, raising copyright‑infringement concerns.

The litigation unfolds amid a broader industry debate about declining digital traffic for news outlets. AI‑generated overviews are frequently blamed for reducing clicks to original reporting, which in turn undermines advertising revenue. Publishers also worry that growing reliance on AI chatbots for news consumption diverts loyal audiences; some analyses indicate small publishers have suffered up to a 60 % traffic loss, with projections suggesting declines exceeding 40 % by 2029.

A statement from Ziff Davis asserts that “OpenAI has copied and monetized Ziff Davis content without permission on a massive scale.” Lance Koonce, partner at Klaris Law and counsel for Ziff Davis, added that since the lawsuit OpenAI has repeatedly misrepresented its ability to search its own datasets for Ziff Davis material and engaged in other serious litigation misconduct.

OpenAI maintains that training AI models on publicly available data constitutes fair use. An OpenAI spokesperson told CNET that as the Times’ case weakens and the newspaper drops certain claims, it continues to pursue “blatantly false allegations” aimed at invading the privacy of unrelated individuals. The spokesperson affirmed the company’s commitment to defending users’ privacy and upholding established fair‑use principles.

In a 2024 rebuttal to the Times’ original suit, OpenAI argued that the newspaper falsely accused it of destroying data and instead accused the Times of secretly deleting its own data that would have demonstrated internal use of OpenAI products. Although the Times has withdrawn one claim against OpenAI, the larger lawsuit remains active.

Other technology firms, including Meta, have faced similar accusations from authors and news publishers regarding copyright infringement. Many of these cases are still pending as courts work to delineate the boundary between fair use and infringement in the era of AI.

Source link