OpenAI's copyright fight with The New York Times: what a new sanctions motion alleges

The New York Times and a group of other news publishers have escalated their long-running copyright lawsuit against OpenAI, filing a new motion for sanctions that accuses the company of hiding evidence relevant to the case. The filing alleges OpenAI concealed tools and datasets capable of identifying copyrighted journalism embedded in the company's training data.

The underlying lawsuit, filed by the Times and joined by other publishers, centers on whether OpenAI's practice of training its large language models on vast amounts of internet text, including copyrighted news articles, without explicit licensing agreements amounts to copyright infringement. OpenAI has argued its use of such material falls under fair use protections.

According to the new filing, publishers allege OpenAI represented to the court that it lacked the technical ability to search its own training data and internal logs for copyrighted material, an assertion the plaintiffs say was untrue. The motion claims the company in fact had tools capable of this kind of search but did not disclose them during earlier stages of discovery.

The publishers further allege that OpenAI held onto billions of internal logs relevant to the case longer than previously disclosed, information that they say could have been used to demonstrate how frequently ChatGPT's outputs closely mirrored specific published news articles rather than producing genuinely original text.

OpenAI has not yet filed a detailed public response to the specific allegations in the new sanctions motion, though the company has consistently maintained throughout the broader litigation that its AI training practices are lawful and that any surface-level similarities between model outputs and news articles do not constitute infringement.

Legal experts following the case say sanctions motions like this one are a serious development in complex litigation, since courts can impose penalties ranging from adverse evidentiary rulings to financial sanctions if they find a party withheld or misrepresented evidence during discovery.

The case is one of several major copyright lawsuits AI companies are currently facing from publishers, authors and other rights holders, all testing similar legal questions about whether training AI models on copyrighted material without explicit permission is protected under existing copyright law.

The outcome of these cases could have significant financial implications for the AI industry more broadly, since a ruling against fair use protections could require companies to negotiate licensing deals for training data at a scale that would be costly and logistically complex to implement retroactively.

Some AI companies, including OpenAI, have already begun striking licensing agreements with individual publishers as a way to secure legitimate access to news content while the broader legal questions remain unresolved in court, a strategy that may also serve to demonstrate good faith to judges overseeing these disputes.

A ruling on the sanctions motion is expected in the coming months, and legal observers say it could shape how much scrutiny courts apply to AI companies' internal data practices in the broader wave of copyright litigation still working its way through the US court system.