TLDR
Unsealed court documents show OpenAI employees knew ChatGPT could bypass the New York Times paywall and celebrated the capability, giving plaintiffs' lawyers what they call direct evidence against the company's fair use defence. Microsoft's CEO has since testified that paywalled content should be licensed. For Australian publishers, the case matters because Australia has no fair use doctrine at all.
KEY TAKEAWAYS
What the documents show
The filing is dated 29 December 2025. OpenAI researcher Nick Ryder alerted company president Greg Brockman to a method to bypass the New York Times paywall. Brockman's reply was positive.[1]
Two other pieces of internal evidence sit alongside that exchange in the plaintiffs' discovery materials. OpenAI's own records show the company knew "the objective" of its networks was to "memorize training data," and that employees were concerned GPT-4 would be "insanely good at regurgitation."[1]
Plaintiffs' counsel put it plainly in the unsealed filing: "OpenAI not only knew that ChatGPT could 'get past a NYT paywall,' but that 'the objective' of its networks was to 'memorize training data' and that employees were concerned that GPT-4 would be 'insanely good at regurgitation.'"[1]
Why this damages the legal defence
Fair use in the United States turns on whether the use of copyrighted material is transformative: does the new work do something fundamentally different with the original, or does it simply reproduce it? OpenAI has argued its models learn patterns and generate new content rather than copy articles wholesale.
Knowing your system was built to memorise training data, and that it could extract paywalled content on demand, makes that argument harder to hold. An employee raising a paywall bypass as a feature rather than a bug reads as the goal, and plaintiffs' lawyers say these documents dismantle the transformative use claim at its foundation.
Microsoft chief executive Satya Nadella widened the gap further when he testified that paywalled content should be licensed and said he would have required retraining had he known models were trained on paywalled content.[1] Nadella is a co-defendant telling a court the opposite of what OpenAI has argued publicly.
What OpenAI says
OpenAI's public response dismisses the paywall evidence as a user behaviour issue. "They claim they might find examples of our users using ChatGPT to try to get around their paywall," the company said in a statement published on its website.[6]
Chief Strategy Officer Jason Kwon said the models were built for something categorically different. "AI models are profoundly transformative. They use massive computational power to learn deep mathematical patterns, analyses, and insights from trillions of datapoints so they can create new content," Kwon said, adding that the systems "have safeguards to avoid replicating the material they learn from."[6]
The Times launched its digital paywall in 2011 and has since grown to 10.1 million paid subscribers and between 50 and 100 million weekly digital users, figures OpenAI's own lawyers filed with the court on 4 April 2025.[2] Those are the numbers attached to the publisher whose paywall the internal messages treated as a problem to route around.
Why Australian publishers hold the stronger hand
The New York Times case is being fought on American fair use grounds. Australia does not have fair use. Section 40 of the Copyright Act 1968 provides fair dealing exceptions for research, criticism, review and news reporting, and nothing broader than that.[3] AI training does not appear on that list.
The Albanese government made that position explicit on 26 October 2025. Attorney-General Michelle Rowland ruled out any Text and Data Mining exception, saying: "This Government has repeatedly said that there are no plans to weaken copyright protections when it comes to AI."[4] Rowland simultaneously announced the government would work through the Copyright and Artificial Intelligence Reference Group (CAIRG), established 5 December 2023, to examine licensing and transparency measures.[5]
For News Corp, Nine Entertainment and independent publishers negotiating with AI companies, that combination leaves no safe harbour to argue around and no text and data mining carve-out to invoke. The unsealed American documents showing what OpenAI employees said when they discovered the models could reach behind a paywall will be read in Sydney boardrooms as carefully as they are in Manhattan courtrooms. CAIRG's next session is already being watched for any shift in the licensing framework.
SOURCES & CITATIONS
- News Plaintiffs' Supplemental Memo on Deletion of Logs, 29 December 2025
- OpenAI court filing, 4 April 2025
- Copyright Act 1968, Section 40, Australian Government
- Attorney-General Rowland announcement on AI and copyright, 26 October 2025
- Copyright and Artificial Intelligence Reference Group (CAIRG)
- OpenAI response to the New York Times
FREQUENTLY ASKED QUESTIONS
What is the 'ah nice' exchange and why does it matter?
What is the difference between fair use and fair dealing?
What is CAIRG and what does it do?
Did Microsoft contradict OpenAI in court?
Xaviery Malinao writes for Prompt the Market on how brands and agencies are adapting to answer engines, drawing on Bushnote's work with clients across search, AI search and content.
