NYT v. OpenAI Cleared to Proceed: The Copyright Case of the Century Heads to the Merits
On March 26, 2025, a federal judge rejected OpenAI's motion to dismiss, letting The New York Times's core copyright claims proceed — with Microsoft as co-defendant.
Per AP and NPR, on March 26, 2025 a New York federal judge denied OpenAI's motion to dismiss, allowing The New York Times and other newspapers to press their core copyright infringement claims against OpenAI and Microsoft, trimming only secondary DMCA claims.
Filed in late 2023, it is legacy media's weightiest copyright challenge to AI: the Times alleges ChatGPT was trained on millions of its articles without authorization and can reproduce paywalled content. Clearing dismissal sends the case into discovery and trial on the merits.
Against Anthropic's $1.5 billion settlement the same year, the NYT case will answer the industry's foundational question: is training on news content fair use?
Why the Anthropic Settlement Is No Precedent Here
The $1.5B Anthropic books settlement (see our coverage) structurally dodged the core question: the court had already found training itself potentially transformative, and the payout covered the acquisition route — downloading pirated book libraries. The NYT case is different: the articles came from the open web, not shadow libraries, so the dispute strikes at training and output directly — above all whether ChatGPT can reproduce paywalled content and substitute for a subscription. The 'market harm' prong of fair use will face its first head-on test in a news context — the hard question every settlement so far has skirted.
Discovery is equally consequential: the Times gains the right to probe training-data composition and internal communications — exactly the information model vendors have withheld as trade secrets. Whatever the verdict, discovery alone will reset the industry's transparency baseline.
Our Take
Whoever wins, the case is rewriting the content-model bargain: a Times victory would crystallize a training-license market overnight, handing pricing power to content owners; an OpenAI victory still leaves publishers having leveraged litigation into a wave of licensing deals — the lawsuit itself is the negotiating chip. In a US with no federal AI statute (see our SB 1047 coverage), copyright litigation is effectively drawing AI's data boundaries in place of regulators; for creators and tool vendors, 'can you trace your training data' is turning from an ethics question into a financial one.
This article aggregates official announcements and public reporting; original sources are linked below.
Source:美联社