Newly disclosed material in The New York Times’ copyright case against OpenAI and Microsoft cites internal discussions about paywalls, publisher displacement and copyrighted training data as the companies continue arguing that AI model training qualifies as fair use.
By yourNEWS Media Newsroom
New court filings in The New York Times’ copyright lawsuit against OpenAI and Microsoft are putting the companies’ own internal communications at the center of the dispute, including allegations that OpenAI employees discussed circumventing the newspaper’s paywall and that Microsoft personnel privately questioned the use of copyrighted material to train artificial intelligence systems.
The newly disclosed material adds detail to a case that could help define how U.S. copyright law applies to the enormous quantities of books, journalism and other material used to develop generative AI.
The Times and other publishers are seeking billions of dollars in damages, according to the Financial Times, arguing that OpenAI and Microsoft copied protected journalism without permission while developing commercial AI products.
The litigation dates to 2023, when the Times sued OpenAI and Microsoft over the alleged unauthorized use of its journalism.
OpenAI disputes the publishers’ legal theory and maintains that training artificial intelligence models on publicly available material is protected by the doctrine of fair use. The company argues that its systems learn patterns from information rather than simply functioning as repositories for copyrighted works and that the resulting models are transformative.
The publishers contend otherwise.
“Defendants repeatedly copied millions of . . . copyrighted articles in their entirety without permission to produce substitutive commercial AI products,” their latest filing states.
Recently disclosed communications are being used by the plaintiffs to argue that personnel inside both companies recognized concerns about obtaining copyrighted material and about AI products competing directly with the publications whose work helped train them.
One allegation involves OpenAI co-founder and President Greg Brockman.
According to the filing, an OpenAI employee told Brockman about “a hack to get around NY Times paywall.”
Brockman allegedly responded, “ah nice.”
The filing presents that exchange as evidence that senior OpenAI leadership was aware of a technique for obtaining material behind the newspaper’s subscription barrier.
The context surrounding that brief exchange remains important. Much of the newly public material is quoted in the publishers’ litigation filings, while some of the underlying exhibits remain sealed, meaning the complete conversations are not publicly available.
The publishers also cited writing attributed to Brockman from approximately 2017 in which he described himself as “deeply motivated by the gazillions” while considering the commercial potential of OpenAI technology.
That statement is being used by the plaintiffs as part of a broader argument that OpenAI’s development ultimately became commercially motivated, an issue potentially relevant to the multifactor legal test courts apply when evaluating fair use.
Another portion of the filing focuses on what OpenAI employees allegedly believed AI would mean for publishers.
The company’s head of ChatGPT is quoted describing generative AI products as an “existential threat” to publishers and writing that such products “are largely substitutive, period [and] will get more and more substitutive as they get better.”
That language could become important because one of the factors courts consider in a fair-use analysis is whether the challenged use harms or substitutes for the market for the original copyrighted work.
OpenAI’s current position remains that its models are transformative and that AI training constitutes fair use. The company also says its products do not improperly substitute for the works used during training and has asked the court for summary judgment on key claims.
Microsoft employees expressed their own concerns, according to the newly disclosed records.
Brent Hecht, Microsoft’s director of applied science, was quoted internally describing large-scale AI training on copyrighted material as “an astonishing theft of unprecedented proportions.”
Hecht also warned that accepting the companies’ fair-use position could “make a complete mockery of the idea of ‘fair use.’”
Microsoft said those remarks should not be treated as the company’s legal position.
“These comments reflect one employee’s individual perspective, are not a legal analysis,” Microsoft said.
The filings also cite Microsoft’s internal measurements of how AI-generated answers affect traffic sent to original publishers.
According to data presented by the plaintiffs, click-through rates to underlying content were between 83% and 93% lower for people using Microsoft’s AI answer tools compared with people using a conventional search engine.
The publishers argue that those figures support their contention that AI systems can increasingly satisfy a user’s information request without sending that person to the website that originally produced the journalism.
Microsoft personnel themselves reportedly examined the potential economic consequences.
Internal material cited in the litigation described the possibility that AI systems could weaken the financial foundation of the publishers and other creators producing the material needed to build and improve large language models.
Microsoft CEO Satya Nadella’s deposition has also become part of the publishers’ case.
Nadella testified that he “had been made aware that OpenAI had scraped and trained on information that was behind a paywall,” according to the filing.
He was also questioned about Microsoft’s contractual ability under certain circumstances to require OpenAI to retrain its artificial intelligence models.
Nadella testified that over the paywall issue he would have required OpenAI “to retrain its models.”
The testimony has been cited by the publishers to argue that even Microsoft’s chief executive recognized a distinction between ordinary publicly available internet material and content protected by subscription barriers.
Microsoft cautioned against interpreting Nadella’s testimony as a legal conclusion about infringement.
The company said his testimony “spoke to broad principles and changes under way in how people find and consume information” and was not a “conclusion about copyright questions.”
The disagreement goes to the central unresolved issue in the litigation: whether copying copyrighted material to train generative AI constitutes a sufficiently transformative use under copyright law or an unauthorized commercial exploitation of protected works.
OpenAI says the former.
On its own account of the litigation, the company says its models are trained using publicly available information and that such training is “grounded in fair use.” It argues that models analyze patterns and relationships across enormous quantities of information to generate new outputs rather than simply reproducing the source material.
The publishers argue that the amount of copying, the commercial purpose of the models and their ability to substitute for original journalism point toward infringement.
According to the plaintiffs, OpenAI copied millions of news articles during model development, including substantial amounts of material produced by the Times and other publishers.
The new filing is intended in part to show that the companies understood the potential economic consequences while continuing to obtain and use journalistic material.
OpenAI has rejected the broader characterization that its technology is fundamentally hostile to journalism. The company says it supports publishers, has entered partnerships with numerous news organizations and believes AI can help newsrooms reach audiences and improve their operations. It has also argued that the Times’ legal claims have narrowed as the case has progressed.
The publishers maintain that those partnerships do not resolve the question of whether copyrighted works were previously copied without licenses.
The latest filings therefore put more attention on what company officials were saying internally while the technology was being developed.
The alleged paywall exchange involving Brockman could bear on how OpenAI acquired some material. The comments attributed to OpenAI’s ChatGPT leadership could bear on whether AI products substitute for publishers. Hecht’s remarks reveal concerns within Microsoft about copyrighted training data, while the company’s click-through measurements provide a quantitative argument about potential effects on publisher traffic.
Nadella’s testimony adds another dimension by addressing what Microsoft says should happen when material behind a paywall is used for AI training.
None of those statements by themselves determines whether copyright infringement occurred. Fair use is a legal determination based on multiple statutory factors and the specific circumstances of the challenged activity, and the federal court has not issued a final ruling resolving the core copyright dispute presented by the publishers.
The newly disclosed evidence nevertheless provides a closer look at internal discussions that occurred while OpenAI and Microsoft were developing technology now central to their businesses.
The Financial Times report describes the dispute as increasingly focused on whether the companies’ own statements undermine their defense that training and operating generative AI are transformative uses permitted under copyright law.
That question — whether the technology legally transforms copyrighted journalism or commercially substitutes for it after mass copying — remains for the court to resolve.
Original article
Comments