The New York Times sues OpenAI and Microsoft over copyright

Background

Before any filing, the two sides tried to settle the question themselves. Talks between The New York Times and OpenAI and Microsoft over copyright began in April 2023, and they failed ^[3]. The lawsuit followed that breakdown.

Copyright law gives a rightsholder control over copying of a work and over certain uses of it. Training a machine-learning model on text involves copying that text into training data and then learning statistical patterns from it, so the question of whether such copying is permitted sits underneath every model built from published writing. In the United States, defendants in copyright cases commonly argue fair use, which asks courts to weigh the purpose of the use, the nature of the work, how much of the work was used, and the effect of the use on the market for it. Those are the questions this case put before a court.

Large language models need very large collections of text, and news archives are attractive material for building them: they are edited, dated and written in full sentences.

What happened

On 27 December 2023 The New York Times sued OpenAI and Microsoft for copyright infringement ^[1][2]. The complaint was filed in the federal district court in Manhattan ^[1][2]. With that filing the Times became the first major US media organisation to sue an AI company for copyright infringement ^[1][2].

The central claim concerned what went into the models. The complaint said the companies used millions of Times articles without permission to build AI products including ChatGPT ^[2]. Copyright infringement, in the ordinary sense, means copying or otherwise using a protected work without the rightsholder's permission and without a legal excuse.

What the Times asked for

The newspaper described its loss in commercial terms. It said the AI tools diverted traffic that would have gone to its website, costing it advertising, licensing and subscription revenue ^[1]. For a publisher, attention to the site is what advertising and subscription sales are priced against, so a product that answers readers' questions without sending them to the site touches revenue directly.

On remedies, the complaint did not claim a specific amount of money. Instead the Times asked the court to destroy datasets containing its content ^[1]. It also sought a jury trial and asked the court to stop the companies from continuing to use its content ^[2]. Destroying training data goes further than paying for past copying: it targets the material a model was built from, not only the act of building it. In a federal civil case in the United States, a jury can be asked to decide disputed facts, while the judge handles the law.

Response and significance

OpenAI said it respects the rights of content creators and is committed to working with them ^[1].

The case drew attention for who brought it and where. It was the first copyright suit against an AI company by a major US media organisation ^[1][2], and it was filed in the federal district court in Manhattan ^[1][2]. It also moved a disagreement that had been conducted in licensing talks into a public courtroom ^[3]. A complaint is one party's account of a dispute: it sets out the Times's claims and requests, not a court's findings. That is what makes the document useful to read — it states, in detail, which uses of published writing a large newspaper considers unacceptable, and what it wants done about them.