Judge orders tougher OpenAI deposition

A federal judge in Manhattan told OpenAI to put up a better witness after the company’s first corporate representative could not answer “even the simplest questions” about an internal anti-plagiarism project called Project Giraffe. The order came in The Intercept’s copyright case in the Southern District of New York on April 7, 2026. (pacermonitor.com) This was not an ordinary employee interview. It was a Rule 30(b)(6) deposition, a federal procedure that lets one side question a company through a person chosen to speak for the company itself on listed topics. (law.cornell.edu, americanbar.org) That rule is built to stop a corporation from playing musical chairs with knowledge. The company can pick any witness it wants, but that witness has to be prepared to give the organization’s answers based on information “known or reasonably available” to the organization. (americanbar.org, markowitzherbold.com) OpenAI chose John Vincent “Vinnie” Monaco to testify about Project Giraffe, which appears to be part of the company’s effort to study whether chatbot outputs reproduced protected text. Judge Ona Wang wrote that Monaco prepared mainly from his own memory, some source code, part of another witness’s transcript, notes tied to researcher Lilian Weng, and “some metrics.” (pacermonitor.com) The judge said that was not enough. Her order says Monaco could not name another person who worked on Project Giraffe beyond a small handful of people he mentioned, and at one point paused for 18 seconds before saying he could not answer whether those people did “most” of the work. (pacermonitor.com) Wang also said OpenAI’s lawyers kept interrupting with objections that “unnecessarily delayed and impeded” the questioning. She ordered another 3.5 hours of deposition time and warned that sanctions could still come later, including money penalties or even having certain answers treated as admissions by OpenAI. (pacermonitor.com) This fight sits inside a much bigger pile of cases. In April 2025, the Judicial Panel on Multidistrict Litigation transferred a large group of OpenAI copyright suits into coordinated pretrial proceedings in the Southern District of New York before Judge Sidney Stein, with discovery issues often handled by Judge Wang. (courtlistener.com) Those cases are not just about what a model says to users. They are also about what data went in, what tests the company ran, what filters it built, what records it kept, and whether it can explain those systems under oath months or years later. (courtlistener.com, pacermonitor.com) Judges in the same litigation have already been forcing OpenAI to hand over unusually large discovery. On January 5, 2026, Bloomberg Law reported that Judge Stein upheld an order requiring OpenAI to produce 20 million anonymized ChatGPT logs to news plaintiffs after rejecting the company’s privacy objections. (news.bloomberglaw.com) Put those two rulings together and the picture gets clearer. The court is not letting OpenAI answer copyright claims with broad assurances; it is demanding documents, logs, and witnesses who can explain how the machine was built and how the company tested whether it copied protected work. (news.bloomberglaw.com, pacermonitor.com) That matters beyond OpenAI because every generative artificial intelligence company facing copyright suits may hit the same wall. If a vendor trained on huge datasets, changed systems quickly, and kept thin records, discovery can turn into a test of whether the company actually knows its own pipeline well enough to defend it in court. (courtlistener.com, law.cornell.edu)