Judge Rips OpenAI Witness

A federal judge in Manhattan said OpenAI’s designated witness was so unprepared that he could not answer “even the simplest questions,” then ordered another 3.5 hours of testimony after a January deposition went off the rails. The witness was John Vincent “Vinnie” Monaco, who OpenAI had put forward to speak for the company on an internal anti-infringement effort called Project Giraffe. (pacermonitor.com) Judge Ona T. Wang wrote on April 7 that Monaco relied on his own “hazy recollections,” a partial transcript, notes that “may or may not” have been the right meeting notes, and “some metrics,” but spoke to no other people to prepare. Wang also wrote that OpenAI’s lawyer made repeated objections that “unnecessarily delayed and impeded” the deposition. (pacermonitor.com) This was not a trial about whether ChatGPT is generally legal. It was a discovery fight inside The Intercept’s lawsuit, a narrower case filed on February 28, 2024 in federal court in New York over whether OpenAI removed copyright management information from journalists’ work. (courtlistener.com) Copyright management information is the label attached to a story: the author name, title, publisher, and copyright notice. The Intercept says OpenAI stripped out that label when using articles, which matters because United States copyright law has a separate rule aimed at people who remove that information. (courthousenews.com) Judge Jed Rakoff kept that one claim alive on February 20, 2025 while dismissing The Intercept’s separate distribution claim against OpenAI and all claims against Microsoft. That meant the case moved from the “is this complaint legally enough” stage into the “show me the documents and witnesses” stage. (law.justia.com) Project Giraffe sits in the middle of the current fight because it was described as an internal OpenAI effort to identify and limit outputs that might reproduce copyrighted text. Plaintiffs wanted Monaco to explain who worked on it, what it measured, and what OpenAI knew about the risk that ChatGPT could spit back protected reporting. (pacermonitor.com) According to Wang’s order, Monaco could not name a single other person who worked on Project Giraffe beyond a small group already mentioned in the deposition, even after saying others were involved. At one point, the judge wrote, he paused for 18 seconds and still could not say whether four named people did “most” of the work. (pacermonitor.com) That matters in a Rule 30(b)(6) deposition, where a company does not get to send someone in to guess. Federal procedure requires the witness to be prepared to give the corporation’s knowledge on the listed topics, because the testimony is treated as the company’s own answer, not just one employee’s memory. (pacermonitor.com) Wang did not punish OpenAI immediately. She said sanctions were deferred until after the added deposition, but warned that any remedy could be monetary or could go further and treat certain answers as admissions by OpenAI itself. (pacermonitor.com) This fight is landing while OpenAI is already losing major discovery battles in related copyright cases in the same New York court. In January 2026, District Judge Sidney H. Stein upheld an order forcing OpenAI to turn over 20 million anonymized ChatGPT logs to news-organization plaintiffs in the broader consolidated litigation. (news.bloomberglaw.com) So the immediate story is not that a judge decided OpenAI infringed journalists’ copyrights. The immediate story is that a federal judge looked at the company’s chosen witness, found the preparation badly lacking, and put OpenAI on notice that the next round of testimony could shape sanctions as well as the facts the court accepts going forward. (pacermonitor.com)