Unsealed court filings show a Microsoft scientist called AI training data collection an astonishing theft

[Jeffy]

The New York Times' ongoing copyright lawsuit against Microsoft and OpenAI has produced a genuinely damaging set of internal statements now that a partly unredacted court brief has become public, and the most striking line in the whole filing comes from inside Microsoft itself. Brent Hecht, the company's director of applied science, described the mass collection of data used to train AI systems as an astonishing theft of unprecedented proportions, going as far as calling it possibly the largest theft of labor in human history, language that reads far more like something from a Times press release than an internal Microsoft document.

Microsoft was not the only company caught making uncomfortably candid internal admissions. OpenAI's head of ChatGPT, Nick Turley, is quoted describing the company's own products as largely substitutive, period, going on to warn that publishers faced what he termed an existential threat from exactly the kind of AI generated answers that increasingly replace a visit to the original news source. Microsoft CEO Satya Nadella himself testified separately that chatbot conversations had substituted for publisher sites, a direct executive level acknowledgment of the same substitution dynamic Turley described from inside OpenAI.

Microsoft's own internal analysis reportedly went further still, documenting a self reinforcing problem the company internally referred to as a doom loop, in which AI generated answers reduce the website traffic that originally supplied the training data those answers depend on in the first place, a genuinely circular problem where the technology's own success gradually starves the ecosystem it was trained on.

The legal stakes riding on these particular admissions are significant. The Times originally filed this lawsuit back in December 2023, alleging unauthorised use of its journalism in training AI systems, and the case has since been folded into broader multidistrict proceedings alongside similar claims from other publishers. The court has already dismissed the contributory infringement and trademark claims specifically, narrowing what remains to direct infringement and the core fair use question, which is exactly the area these newly unsealed internal statements are aimed squarely at. The Times is using Hecht's and Turley's own words to argue that both companies clearly understood their products could function as direct market substitutes for the very material they trained on, a factor that weighs heavily against a fair use defence under the purpose and market effect prongs of that legal test.

Both companies are pushing back on how much weight these statements should actually carry. Microsoft maintains that Hecht's comments represented his own personal perspective rather than any kind of official corporate legal position, while OpenAI has largely stayed quiet on the specific quotes and continues to maintain more broadly that AI training itself qualifies as fair use regardless of what individual employees may have said internally. Whether a judge finds an executive level and a senior scientist level admission this candid more persuasive than that formal legal defence is likely to be one of the more consequential questions in the entire AI copyright litigation landscape currently working through American courts.

Key facts: Microsoft tied to news suit; Says fair use; Posts AI codewww.tradingview.com

[Lowkey]

Hecht calling it possibly the largest theft of labor in human history from inside Microsoft's own applied science division is the kind of quote that plaintiffs' lawyers dream about finding in discovery. That is about as damaging an internal admission as you could hope to unearth in a copyright case like this one.

[QuantumFoam19]

Microsoft trying to distance itself from Hecht's comments as merely personal opinion rather than official position is a predictable legal move but it is going to be a hard sell once a jury or judge hears a senior applied science director calling his own company's data practices the largest theft of labor in human history.

[Aura49]

The doom loop framing Microsoft used internally is honestly the more interesting long term problem buried in this story. If AI answers really are reducing the traffic that funds the journalism those answers were trained on in the first place, that is a genuinely unsustainable cycle regardless of how the lawsuit itself gets resolved.

[Amber81]

Nadella's own testimony about chatbot conversations substituting for publisher sites is arguably more damaging than either of the other two quotes precisely because it comes directly from the CEO rather than an individual researcher whose views can be dismissed as personal. That is about as close to an official corporate acknowledgment as this case is likely to get.

[Rachel83]

The purpose and market effect prongs of fair use are exactly where these admissions land hardest, and it makes sense the Times is building its whole argument around them. If the companies genuinely understood their own products would substitute for and undercut the original market, that undermines the fair use defense in a pretty fundamental way.

[NatureBoyDave24]

Turley's largely substitutive, period quote is such a blunt admission it barely needs any spin from opposing counsel, it is already doing exactly the work the Times wants it to do sitting there entirely on its own without any additional argument needed around it.

[CacheLayerShark]

Nearly three years since the original December 2023 filing and we are still working through discovery and unsealed briefs rather than an actual verdict tells you how slowly this entire category of AI copyright litigation is moving, even though it is arguably one of the most consequential legal questions facing the entire industry right now