News Plaintiffs' unsealed public summary-judgment brief quotes Microsoft's own applied-science director calling AI training "the largest theft of labor in human history," plus a sworn Nadella admission of substitution

What happened

The News Plaintiffs (New York Times, Daily News, Center for Investigative Reporting, Intercept, Ziff Davis) filed the public, redacted version of their Combined Summary Judgment Brief in the same MDL this thread already tracks (Doc. 1977-1, filed 2026-09-17, 92 pages) — read directly from the docket PDF, not from coverage — and it quotes internal Microsoft and OpenAI admissions that go well beyond the fair-use framing either company has used publicly. The brief's opening line: "This case is about, as Microsoft's Director of Applied Science put it, 'an astonishing theft of unprecedented proportions'... perhaps the 'largest theft of labor in human history.'" That Microsoft executive is named later in the same brief as Dr. Brent Hecht, in an internal document he wrote "soon after this lawsuit was filed" (the underlying NYT suit predates this MDL). A separate Microsoft internal document, quoted verbatim, states: "Our AI content strategy has started a 'doom loop' that will hurt the performance of our models and the entire web at the same time: It is highly unusual that an end-product threatens the economic foundations of its essential suppliers, but that is the situation we have created for our LLM business." The brief cites Microsoft's own data showing Copilot's "answer engine" cut click-through rates 83–93% for NYT/Daily News domains and 51–94% for Ziff Davis domains, versus traditional Bing search.
CEO-level admissions under oath are also in the brief: Satya Nadella agreed under oath that conversing with chatbots "has substituted... giving you the information right there on the website on the AI platform versus needing to go to the underlying source." OpenAI's Head of ChatGPT, Nick Turley, wrote that the products are "largely substitutive, period" and "will get more and more substitutive as they get better"; Greg Brockman wrote that the models are "particularly good at predicting text of news articles" and "excellent at news." The brief also describes OpenAI building a "Bloom filter" that suppressed model output only of content belonging to entities that had sued it — not a general anti-infringement measure — which Hecht is quoted calling an "accidental cover up" that would leave "people who have a right over the content [with] less visibility into what was used for training." This filing is the public unsealing of exhibits behind the summary-judgment motions this thread already logged for 2026-09-04; it does not resolve the case, but it puts internal admissions on the public docket that materially undercut the fair-use defense both companies have argued. (CourtListener docket, 1:25-md-03143, Doc. 1977-1; TechCrunch)
San Francisco County Superior Court Judge Harold Kahn overruled most of Anthropic's motion to dismiss Reddit's data-scraping lawsuit, in an order dated 09-17 and filed 09-18 — letting three of Reddit's five claims proceed (breach of contract, interference with contract, and violation of California's unfair-competition law), while dismissing two (unjust enrichment, trespass to chattels) as too vague, with leave for Reddit to refile by October 16. Kahn rejected Anthropic's argument that Reddit's user agreement was an unenforceable "browsewrap" contract, pointing to Reddit's allegation that Anthropic kept scraping the site more than 100,000 times even after Reddit's CEO publicly called out the behavior. Reddit's suit — filed June 2025, alleging Anthropic used automated bots to scrape the platform and trained Claude on users' personal data without consent — pleads contract and business-tort claims, not copyright infringement, so this is adjacent to rather than part of this thread's copyright-litigation core; filed here for completeness of Anthropic's training-data legal exposure, the same basis the 08-28 Pentagon ruling was logged on. (Yahoo News/Mashable)

Sources 3

CourtListener docket, 1:25-md-03143, Doc. 1977-1 courtlistener.com primary source
TechCrunch techcrunch.com high 0.886 no consensus
Yahoo News/Mashable yahoo.com solid 0.692

Credibility ratings combine two public datasets: Lin et al. 2023, PNAS Nexus 2(9):pgad286 (CC BY 4.0) for the domain-quality score, and Wikipedia's perennial-sources list (CC BY-SA 4.0) for the reliability tier. A rating describes an outlet's general record, never this particular article. Unrated means absent from those datasets — not a judgement.

The wider story

This event sits inside Copyright Exposure, which carries how it developed before and since.

Story 10 of 14 in this thread's timeline, oldest to newest

Adjacent threads (shared entities):

· Anthropic Rents the Buildout

· The Enterprise Agent Land Grab