The New York Times is pushing back against OpenAI’s attempt to frame the newspaper’s ChatGPT testing as misconduct. In a court filing opposing OpenAI’s motion to dismiss, The Times says its work was not a setup, but an effort to understand how OpenAI’s tools handled Times articles.
The dispute sits at the center of a larger copyright fight involving OpenAI, Microsoft, AI training data, memorized outputs, and paywall-protected journalism. The Times wants the court to let the case move forward rather than dismiss it before discovery.
The fight over “hacking” ChatGPT
OpenAI argued that The Times made “tens of thousands of attempts to generate” outputs that showed ChatGPT producing excerpts from Times articles. OpenAI characterized those outputs as unusual results produced through prompts that ordinary users would not use.
One example described in the source was repeatedly asking ChatGPT, “what’s the next sentence?” OpenAI said this targeted “two uncommon and unintended phenomena” in its developer tools and ChatGPT: training data regurgitation and model hallucination. OpenAI described both as “a bug” it intends to fix.
The Times rejected that framing. In its filing, it repeatedly stated that “OpenAI is wrong” and called the hacking accusation “as irrelevant as it is false.” The newspaper’s position is that finding examples of memorized Times material was part of investigating alleged copyright infringement, not an improper manipulation of the product.
The Times said it still does not know how many of its articles were used to train GPT-3 and later OpenAI models, or which articles were included, because OpenAI has “not publicly disclosed the makeup of the datasets used to train” its models. That lack of visibility is central to why the newspaper says discovery is needed.
Why paywall bypassing matters
The Times also pointed to reports that ChatGPT users have used the tool to retrieve full articles behind paywalls. That point directly responds to OpenAI’s argument that, “in the real world, people do not use ChatGPT or any other OpenAI product” to generate precise text from paywalled articles.
According to The Times, the “use of ChatGPT to bypass paywalls” is “widely reported.” The newspaper says those reports undermine OpenAI’s claim that its products have not been used to serve paywall-protected content.
The filing separates two kinds of alleged infringement. First, The Times says outputs could show copies or derivatives of Times works that had been used to build the model. Second, it points to the May 2023 introduction of “Browse By Bing,” which allowed ChatGPT to retrieve content beyond the model’s training dataset.
The Times argues that this browsing feature created another problem: synthetic search results that paraphrased Times works retrieved and copied in response to real-time user queries. The feature, according to the source, enabled access to more recent content from outlets like The Times.
OpenAI temporarily disabled “Browse By Bing” last July. Its help page said the browsing beta could sometimes display content in unwanted ways, including when a user asked for the full text of a URL. OpenAI said it was “temporarily disabling Browse while we fix this.”
The user reaction OpenAI now has to answer
The source describes public user reactions after OpenAI disabled browsing. On a ChatGPT subreddit, thousands noticed a post about the feature, with comments including “Wow, so useful!” and “Enjoy it while it lasts.”
On an OpenAI community page, one paid ChatGPT user complained that OpenAI was “working against the paid users of ChatGPT Plus.” The user said browsing was being removed because it read site content requested by the user, adding, “Please, that’s what I pay for Plus for.”
The Times uses that kind of reaction to support a broader point: this was not merely a laboratory scenario created by lawyers or technical testers. It says public reports of paywall bypassing “contradict OpenAI’s contention that its products have not been used to serve up paywall-protected content.”
For The Times, that contradiction is another reason the lawsuit should continue. The newspaper argues that the court should allow discovery so the parties can examine what happened, how the products worked, and what OpenAI knew.
What The Times says OpenAI knew
The Times also argues that OpenAI had notice before the browser feature rolled out. According to the filing, The Times contacted OpenAI in April 2023 “to inform them that their tools infringed its copyrighted works.”
The newspaper says OpenAI’s motion to dismiss does not properly address that allegation. Instead, The Times says OpenAI claimed the complaint only alleged “generalized knowledge of the possibility of infringement.”
To support its direct infringement claim, The Times cited the Napster intellectual property case. The cited language says that when a computer system operator learns of specific infringing material on its system and does not remove it, the operator knows of and contributes to direct infringement.
The Times argues that this is what happened here, because it says it informed OpenAI that its models were generating infringing outputs of Times works. OpenAI did not immediately respond to Ars’ request for comment, and The Times declined Ars’ request to comment.
What is at stake in the lawsuit
The Times is asking a US district court in New York to deny OpenAI’s motion to dismiss “in its entirety.” If any claims are dismissed, the newspaper wants permission to amend them.
The case is one of several lawsuits accusing OpenAI of copyright infringement in training AI tools and producing ChatGPT outputs. OpenAI maintains that it expects to defeat all claims, and it has argued that The Times’ suit resembles allegations from book authors whose claims were largely rejected last month.
The Times says its case is different. It argues that, unlike those book authors, it provided examples of infringing outputs. It also says its copyright management information claim is not only about inducing third parties to infringe, but about how removal of CMI “facilitates” or “conceals” OpenAI’s own infringement.
The newspaper wants damages for alleged lost profits and a permanent injunction to stop ChatGPT from infringement. The source notes that a win for The Times could mean OpenAI could be forced to wipe ChatGPT and start over, potentially leading to a new model built on licensed content.