What does the New York Times lawsuit against OpenAI mean for AI and copyright?

This week brought the most closely watched copyright case yet over generative AI. On Wednesday, December 27, The New York Times Company filed a complaint in federal court in Manhattan against Microsoft and several OpenAI companies. Authors and artists have already sued AI developers, but this is the first such challenge from a major news organization, and it puts a large, well-documented archive at the center of the debate about how AI models are trained.

A lawsuit begins with one side's version of events. A complaint is the document that starts a case: it sets out what the plaintiff (the party suing) says happened and what it wants. The defendants answer later, in court. Nothing in a complaint has been proven, and no judge has ruled on any part of this one.

What the Times claims

The complaint makes several connected arguments.

• Training. It says the large language models behind ChatGPT and Microsoft's Bing Chat (now branded Copilot) were built by copying millions of Times articles, investigations, reviews and guides, without permission or payment, and that Times content was given particular weight in training data.
• Output. It says the tools can reproduce Times articles nearly word for word, summarize them closely and imitate their style, and it attaches many side-by-side examples as an exhibit.
• Search. It says chat answers drawing on Bing search results can deliver long excerpts and detailed summaries, including the product picks of its review site Wirecutter, so readers have less reason to visit the original.
• False attribution. It says the tools sometimes invent information and present it as coming from the Times, which it argues harms its reputation.

The complaint lists seven claims, including copyright infringement, removing copyright information, unfair competition and trademark dilution. It asks for damages it says reach billions of dollars, without naming a figure, a court order stopping the conduct, and the destruction of models and training sets that contain Times works. It also says the Times tried for months, starting in April, to negotiate a licensing deal before suing.

What OpenAI and Microsoft say

Microsoft declined to comment when the suit was filed. OpenAI issued a statement through a spokesperson. It said, "We respect the rights of content creators and owners and are committed to working with them," described its talks with the Times as productive, and said it was "surprised and disappointed" by the lawsuit. It added that it hoped to find a way of working together that benefits both sides, as it says it is doing with other publishers. This year it signed agreements with the Associated Press and with the German publisher Axel Springer.

OpenAI's legal position on training is older than this case. In a 2019 submission to the U.S. Patent and Trademark Office, it argued that, under current law, training AI systems on copyrighted works counts as fair use. The Times's complaint acknowledges that the companies hold this view publicly, and disputes it.

Reading the two sides side by side

A simple way to follow the case is to line up each claim with the opposing argument.

What it could mean for users

For now, nothing changes in how ChatGPT or Copilot works. Cases like this usually take a long time, and they can end in a settlement rather than a ruling. Still, the outcome could shape whether AI companies must license the text they train on, how much that costs, and whether tools are built to refuse requests to reproduce articles. It is also a reminder of a basic habit: when a chatbot attributes a claim to a named publication, check the original.

What is the fair-use question at its heart?

Fair use is a part of U.S. copyright law that permits some uses of protected work without permission. Courts weigh four factors: why and how the work is used, including whether the use is commercial and whether it is "transformative" (adds a new purpose or meaning); the nature of the original work; how much of it was used; and the effect of the use on the market for the original.

OpenAI's argument, as set out in 2019, focuses on the first and fourth factors. It says training copies works only so a program can learn patterns in language, not so anyone can read them, which makes the use highly transformative. It points to earlier decisions, such as the Google Books case, where copying as a step in computer analysis was found to be fair.

The Times argues the reverse on the same factors. It says there is nothing transformative about building a product that competes with the source, and that outputs which recite or closely summarize its articles show the market harm directly. That is the core dispute: whether training is a new kind of use that leaves the original market alone, or a way of replacing it. Only a court can answer that, and this one has not yet begun to try.