Unsealed NYT v. OpenAI filing shows Microsoft, OpenAI executives privately called AI training 'theft' and warned of a web 'doom loop'
2026-09-17
A court filing unsealed Thursday Sep 17, 2026 in the New York Times v. OpenAI/Microsoft copyright suit — a New York Times motion for summary judgment — quotes previously sealed or redacted internal documents and depositions. An internal Microsoft document called LLM training-data use 'an astonishing theft of unprecedented proportions' affecting people never compensated for their content, and separately warned that Microsoft's 'AI content strategy has started a doom loop that will hurt the performance of our models and the entire web at the same time.' Microsoft CEO Satya Nadella testified under oath that clicks from Bing to the New York Times and other news sites fell more than 90% after Bing began surfacing AI-generated content from those sites. Documents showed OpenAI built 'a hack to get around nytimes paywall,' to which OpenAI cofounder Greg Brockman replied 'ah, nice.' Microsoft researcher Brent Hecht wrote that LLMs use content 'without ways of distributing economic value down the supply chain,' threatening creators' economic stability, and OpenAI policy director Jack Clark wrote the company was 'creating systems that substitute for the labor of the people that define the culture of society.' The unredacted filing was surfaced by Jason Kint, CEO of Digital Content Next. 404 Media, which found the filing, and TechCrunch both covered the documents; 404 Media's own framing calls the admissions 'some of the most damning indictments' of how LLMs were trained.
Significance 4: Sworn testimony and internal documents from Microsoft and OpenAI executives, unsealed in active litigation rather than voluntarily disclosed, directly admit to training-data 'theft' and a self-inflicted 'doom loop' harming publishers — a major stated-vs-revealed disclosure with real litigation and reputational exposure; held at 4 rather than 5 since no court ruling has yet resulted.
Implications · machine-drafted, not owner judgment
This strengthens the New York Times' summary-judgment position and hands other publisher plaintiffs suing OpenAI and Microsoft ready-made quotes for amended complaints, raising the industry's aggregate copyright-liability exposure. It also validates the 'doom loop' argument — that generative answers cannibalize the same publisher ecosystem the models are trained on — as a structural risk to the web's content economy, not just a talking point from critics. For the owner's leader-pattern-predicts-behavior thesis, this is a clean stated-vs-revealed gap: public 'transformative use' framing versus private acknowledgment of theft, extending the same pattern already tracked this month for Meta, Anthropic, and OpenAI.
- A ruling on the New York Times' summary judgment motion
- Other publisher plaintiffs citing these same quotes in new or amended filings
- Any public damage-control statement from Microsoft or OpenAI
Sources
- press 'Doom Loop': OpenAI and Microsoft Admits LLMs Are Destroying the Web and Built on Theft retrieved 2026-09-18
- press Microsoft exec called AI scraping 'the largest theft of labor in human history,' new unredacted filings reveal retrieved 2026-09-18