Security1 publisher3 min readPublished
Nadella testified that chatbots have "substituted" for publisher sites
The news publishers' partly unsealed summary judgment motion quotes Microsoft's CEO on substitution and on licensing paywalled training data. Microsoft says he was speaking to broad principles about how people find information, and that his words are not conclusions about copyright.
The Watch · Security desk
What happened
- Microsoft CEO Satya Nadella testified during discovery that conversations with chatbots have "substituted" for publisher sites, according to the news publishers' partly unsealed summary judgment motion.
- Nadella also testified that paywalled content should be licensed for AI training, per the same motion, which was filed Thursday in the consolidated New York case.
- Had he known OpenAI scraped and trained on paywalled information, Nadella testified, he would have invoked Microsoft's contractual right to require OpenAI to retrain its models.
- OpenAI senior executive Nick Turley wrote that "[o]ur products are largely substitutive, period," one of several internal messages the publishers quote in the brief.
Compiled by The WatchSomething wrong?How this is made
Why it matters
- contradiction Microsoft's filings argue Copilot does not substitute for publishers' journalism, while the deposition and internal message the plaintiffs chose to quote both use that word, so the trial fight is now over what a company employee meant, not over whether the sentence exists.
- exposure Publishers who sue later can point at defendant telemetry for market harm instead of commissioning their own traffic study, and the range Microsoft itself produced is wide enough to survive argument at either end.
- decision Buyers negotiating model access now have a named term to ask for, because Microsoft's CEO described holding a right to require its supplier to retrain on a provenance breach.
- precedent Once the confidentiality designation comes off part of a discovery record this large, the quotes are reusable by every other plaintiff and regulator looking at the same training runs.
Summary judgment decides which claims reach a jury. The brief carrying this testimony is the publishers' own, filed in redacted form about two weeks before the court let part of the record out, in the multidistrict litigation before the US District Court for the Southern District of New York [5]. That litigation consolidated dozens of suits accusing the two companies of downloading pirated works, training large language models on protected content, and producing outputs similar to the originals [19].
Microsoft disputes the reading. Spokesperson Alex Haurek said Nadella's testimony and the company's position in the case "are perfectly consistent" [6]. "He spoke to broad principles and changes underway in how people find and consume information," Haurek said. "Those observations should not be confused with conclusions about copyright questions before the Court, which Microsoft addresses in its filings." [7] The company's filings argue that training is fair use in part because it is transformative, and that Copilot does not substitute for publishers' journalism [8].
Brent Hecht, Microsoft's director of applied science, said copying millions of news articles without publishers' permission could be the "largest theft of labor in human history" [9]. Haurek said Hecht's comments "reflect one employee's individual perspective, are not a legal analysis, and do not represent the company's views" [10]. OpenAI did not immediately respond to Bloomberg Law's request for comment [20].
Immediately after the publishers sued, OpenAI built a filter to suppress output of their content, and did not suppress content from entities that had not sued, according to the motion [12]. Hecht called that an "accidental cover up" because it would result in "people who have a right over the content having less visibility into what was used for training" [13]. Per-publisher output suppression was implementable. A lawsuit was what triggered it.
Microsoft's own data supplies the market-harm evidence in the brief: click-through rates for Times and Daily News content down 83% to 93%, and Ziff Davis domains down 51% to 94% [14]. Read from the other end, between 7% and 17% of the two newspapers' prior click-throughs remain, and between 6% and 49% of Ziff Davis's [15]. The publishers call the dynamic a "doom loop" that hurts the market for creative works [23].
The brief also describes content crossing between the two companies. They did what they themselves called "horse trading," supplying copies of articles to each other and on occasion selling content to each other [16], and a Microsoft project code-named Project Taxi gave OpenAI a compilation of "billions" of webpages gathered to support Bing [17]. When staff told OpenAI president Greg Brockman about a hack to get around the Times' paywall to help with scraping efforts, Brockman responded, "ah nice" [22].
The Trump administration urged the court to rule that training AI on copyrighted works constitutes fair use [11]. Steven Lieberman of Rothwell Figg Ernst & Manbeck, counsel for plaintiff New York Daily News, said, "Throughout this case Defendants insisted that these documents be treated as confidential so that the public could not see them." He added: "Well, now the cat is out of the bag." [18]
What to watch
- Whether the SDNY court grants any part of the motions and which training and output claims survive to trial.
- Whether the remaining redactions in the publishers' brief come off, including the second project the motion names alongside Project Taxi.
- Whether the court gives weight to the Trump administration's position that training on copyrighted works is fair use.