Judge tosses OpenAI's defense in NYT copyright battle

Imagine a newspaper publishes an article about a hot tech startup scraping the web. A couple of years later, that startup launches a tool that spits out exact copies of the newspaper's reporting. When sued for copyright infringement, the startup fires back: you wrote about our web crawling years ago, so you missed your deadline to sue.

That was essentially OpenAI's defense against The New York Times. A federal judge was not impressed.

Turning a reporter's scoop into a shield

The legal battle began in December 2023, when The New York Times sued OpenAI, alleging that ChatGPT outputs violate its copyrights by directly regurgitating original news articles.

Instead of just defending fair use, OpenAI tried to get the case thrown out on procedural grounds, claiming the newspaper's claims were time-barred.

OpenAI's star evidence was a single article published by the Times itself in November 2020. In that piece, a Times journalist reported that OpenAI was analyzing a trillion words across the internet.

OpenAI argued that as a sophisticated publisher, the Times should have known back in 2020 that its articles were inside the training batch. It also claimed that other reports made this common knowledge at the time, giving the paper a duty to act immediately.

The judge calls out a straw man

On Friday, US District Judge Sidney Stein officially denied OpenAI's motion to dismiss.

In his ruling, Stein made it clear that the burden of proof rests entirely on OpenAI. The AI maker had to prove the Times actually knew ChatGPT would infringe on its copyrights two full years before the chatbot was even launched in November 2022. OpenAI failed to provide that proof.

Stein noted that a reporter writing about AI training does not mean the publisher knew the tool would eventually regurgitate full articles verbatim.

He also dismissed OpenAI's sophisticated publisher claim, explicitly calling it a straw man argument. The judge wrote that general coverage of web scraping fails to put plaintiffs on notice of the particular infringing conduct at the heart of the lawsuit: the specific copying of their proprietary work.

Why it matters

OpenAI wanted a quick procedural exit, and the court just slammed the door shut. This decision forces OpenAI to fight the real battle on the merits: whether training frontier models on copyrighted journalism and generating near-identical outputs crosses the legal line.

We think OpenAI's argument looked desperate from the start. Conflating public knowledge of bulk data scraping with actual notice of copyright infringement makes little legal sense. If the court had accepted that logic, every AI lab could announce a massive crawl and start the statute-of-limitations clock ticking against the entire internet before anyone even saw the product.

For publishers and independent creators, this is a reassuring precedent. The court established that you do not need to guess what an unreleased model might do behind closed doors. The right to protect your work begins when the infringement actually lands in the wild.