Judge criticises OpenAI witness in copyright case

A federal judge in Manhattan ordered OpenAI to put one of its own witnesses back in the chair after saying he could not answer “even the simplest questions” in a January 27 deposition about how the company tries to stop chatbots from reproducing copyrighted work. The judge gave plaintiffs another 3.5 hours with the witness on April 8 and said sanctions could still come later. (law.justia.com) (pacermonitor.com) The witness was John Vincent “Vinnie” Monaco, and Judge Ona Wang said he prepared by relying mostly on his own memory, some source code he helped write, part of another deposition transcript, notes from OpenAI researcher Lilian Weng, and “some metrics.” Wang wrote that Monaco spoke to no other people before testifying for the company. (law.justia.com) (pacermonitor.com) The project at the center of the fight is called Project Giraffe, which appears to be OpenAI’s internal work on output controls, meaning tools meant to reduce cases where a chatbot spits back text too close to what it saw before. In court, Monaco could not name a single other person who worked on that project beyond a few colleagues already mentioned in questioning. (law.justia.com) (chicagotribune.com) That matters because this was not an ordinary employee interview. Monaco was testifying as a Federal Rule of Civil Procedure 30(b)(6) witness, which is the person a company designates to speak for the company itself on listed topics, so weak answers can bind the corporation the way a bad map can send the whole convoy in the wrong direction. (law.justia.com) (pacermonitor.com) Judge Wang was also angry at the lawyers in the room. Her April 7 order says defense counsel’s objections “unnecessarily delayed and impeded” the January deposition, and she left open penalties that could include money sanctions or a recommendation that certain answers be treated as admissions by OpenAI. (law.justia.com) (pacermonitor.com) This dispute sits inside a much larger pile of lawsuits in the Southern District of New York. The Authors Guild sued OpenAI on September 19, 2023, The New York Times sued OpenAI and Microsoft on December 27, 2023, and The Center for Investigative Reporting sued on June 27, 2024, all accusing the companies of using copyrighted writing without permission in building generative artificial intelligence systems. (authorsguild.org) (courtlistener.com 1) (courtlistener.com 2) The news organizations’ version of the case is not just “you trained on our archives.” The Chicago Tribune report says the plaintiffs accuse OpenAI of stealing and distorting journalists’ work, then serving users incomplete or inaccurate rewrites through chatbots that more people now use to get news. (chicagotribune.com) OpenAI has already won some arguments and lost others in these cases, so this deposition fight does not decide the copyright question by itself. What it does decide is whether the judge thinks OpenAI is meeting basic discovery duties, which is the court process where each side has to hand over information and produce prepared witnesses instead of sending someone in to guess from memory. (ipwatchdog.com) (law.justia.com) By April 7, the judge had already warned that January’s performance was not good enough, and by April 8 the parties had agreed to the extra deposition time she demanded. If Monaco’s second round fixes the record, OpenAI may escape with embarrassment; if it does not, the court has already sketched out the next step in plain language: sanctions, admissions, or both. (law.justia.com) (pacermonitor.com)