US media outlets have accused OpenAI of concealing and destroying evidence in a case involving the use of copyrighted news materials to train ChatGPT. This was reported by Qazaqyia.kz citing Kursiv Media.
The New York Times, Daily News, and several other publishers filed a motion in the federal court in Manhattan, according to Euronews.
The plaintiffs allege that OpenAI and its partner Microsoft are obstructing the disclosure of key evidence that could show how millions of copyrighted journalistic materials were used to train artificial intelligence models.
According to the media companies, OpenAI failed to provide the requested datasets and ChatGPT logs, and misled the court regarding the ability to search for copyrighted content in training data. The publishers believe that recent testimony from a company employee contradicts its previous statements.
New York Daily News attorney Steven Lieberman stated that OpenAI had concealed information about how the language model was trained for two years.
"This motion asks the court to punish OpenAI for concealing and destroying evidence of how ChatGPT was trained on stolen journalism," Lieberman said.
The publishers are asking the court to impose sanctions on OpenAI for alleged violations during the discovery process and to require the company to cover legal costs incurred in attempts to obtain allegedly wrongfully withheld materials.
The New York Times lawsuit against OpenAI and Microsoft was filed in late 2023. The newspaper claims the companies use its journalistic materials without permission or compensation, creating products that compete with news outlets for audience and internet traffic.
According to publishers, the development of AI services has increased pressure on the industry. In particular, after the introduction of AI overviews in Google's search engine, users began to visit original source websites less frequently, negatively impacting editorial advertising revenues.
Other media companies later joined The New York Times lawsuit, including the publisher of Daily News and Chicago Tribune MediaNews Group, digital holding Ziff Davis, and the nonprofit Center for Investigative Reporting.
In turn, OpenAI and other AI developers argue that training models on books, articles, and other content published online falls under the fair use principle under US law. This issue is currently being considered in dozens of lawsuits involving book authors, artists, musicians, and other rights holders.
The largest settlement in such disputes to date has been the agreement between Anthropic and book authors. Under it, the developer of Claude agreed to pay rights holders $1.5 billion for using works to train the model.
According to disclosed reports, The New York Times has already spent more than $28 million on litigation with AI companies, including a separate lawsuit against Perplexity.
Despite ongoing legal disputes, more media companies are choosing to enter into licensing agreements with AI developers. OpenAI, Google, and Meta pay a number of publishers for the right to use their news materials and archives to train their models.