Doom Loop: Unsealed Filing Hits OpenAI and Microsoft
An unsealed summary judgment filing shows both companies warned internally that their own models were wrecking the web they depend on.
In short
Yes — a court filing unsealed on September 17, 2026 quotes Microsoft's and OpenAI's own internal documents warning that their models set off a “doom loop” that strips the web of the content those models are trained on.
At a glance
- The source is a summary judgment motion filed by New York Times lawyers in the U.S. District Court for the Southern District of New York.
- Unsealed on September 17, 2026; Jason Kint of Digital Content Next surfaced the unredacted version.
- An internal Microsoft document calls the practice the largest theft of labor in human history.
- Testimony from Microsoft executives puts the drop in clicks to news sites on Bing at more than 90 percent.
- OpenAI policy director Jack Clark wrote of systems that substitute for culture-defining work.
The most damaging language about generative AI's effect on the web now comes from inside the companies building it. A filing unsealed on September 17, 2026 in the New York Times copyright case quotes Microsoft and OpenAI documents describing a self-inflicted “doom loop” and “the largest theft of labor in human history.”
What the document actually is
The quotes sit inside a motion for summary judgment written by lawyers for the New York Times and filed in the U.S. District Court for the Southern District of New York. Jason Kint, who runs the trade group Digital Content Next, found and publicized the unredacted version. 404 Media went through the filing on September 17, 2026.
The admissions
One internal Microsoft document reportedly frames training on other people's work as “an astonishing theft of unprecedented proportions.” The same material states that the company's own AI content strategy started a doom loop that damages the content supply chain and, downstream, model quality itself. Microsoft executive Brent Hecht wrote that without a way to push economic value back down the supply chain, the financial footing of the people producing content is put at risk. A separate Microsoft policy paper cautioned that generative AI could badly disrupt employment for the very people whose data trained it.
On the OpenAI side, the filing cites policy director Jack Clark writing about building systems that substitute for the labor of people who define a society's culture. It also describes an internal workaround for the New York Times paywall, to which co-founder Greg Brockman replied “ah, nice.” An OpenAI software engineer testified that users simply do not click source links, however prominently they are displayed.
Why the traffic number carries the argument
Testimony from Microsoft executives, including chief executive Satya Nadella, puts a figure on the damage: clicks to news sites on Bing fell by more than 90 percent once their material started appearing in answers. That is the loop in one line — summarize the source, remove the visit that pays for it. Fewer paid newsrooms means less new reporting, and less new reporting means thinner training data for the next model. The filing also has OpenAI describing itself as an existential threat to news publishers.
What this does not establish
These are excerpts selected by one side of an active lawsuit. No court has ruled on the motion, and the surrounding context for each internal document comes from the filing rather than from the defendants. The report reviewed here carries no response from OpenAI or Microsoft to the quoted passages. A second piece on the same filing at The Verge could not be retrieved for this article, so any additional framing it offers — including the “Google Zero” angle — is unverified here and deliberately left out.
FAQ
What is the AI doom loop?
Per the quoted documents, it is a feedback cycle: AI answers replace visits to the original source, the source loses revenue, less content gets produced, and the models lose the training data they rely on.
Where do the doom loop quotes come from?
From a summary judgment motion filed by New York Times lawyers in the Southern District of New York, unsealed on September 17, 2026, which reproduces internal Microsoft and OpenAI documents.
Have OpenAI or Microsoft responded to the filing?
The report reviewed here includes no comment from either company on the quoted passages, and the court has not ruled on the motion.
