According to Bloomberg, OpenAI’s former chief technology officer Mira Murati said on Thursday that OpenAI would likely have disintegrated if Sam Altman had not returned to the CEO position after being briefly ousted in 2023. Her comments were the clearest explanation yet of one of Silicon Valley's most dramatic boardroom power struggles.
Mulati later left OpenAI and founded Thinking Machines Lab. She was one of the key figures in pushing for Ultraman's reinstatement after he was abruptly fired nearly three years ago. That experience thrust Mulati into the spotlight: She was briefly named interim CEO. As she quickly moved to try to bring Ultraman back, the board replaced her with former Twitch CEO Emmett Shear.
At that time, more than 700 of OpenAI's approximately 770 employees signed a joint letter threatening to resign and follow Altman to Microsoft. Just five days after Altman was fired, he was reinstated with the support of a reconstituted board.
"When I realized that their decision could have catastrophic consequences for the company and that it could even fall apart, I felt I had to act quickly," Mulati told the Bloomberg Technology Conference in San Francisco on Thursday. "If I hadn't done that, OpenAI might have disintegrated."
Mulati's remarks echoed her testimony in April this year in Elon Musk's case against OpenAI. She said at the time that the company was at risk of falling apart after Altman was ousted. She also admitted in her testimony that she had concerns about Altman's leadership style. Helen Toner, a former OpenAI board member, gave a more complex description of Mulati's role in a deposition, saying that Mulati "was watching the wind, without realizing that she was the one deciding which way the wind was blowing."
"Even though on the surface it all looked very chaotic, at every point I knew exactly what I needed to do," Mulati said. "That was to provide continuity and stability, to help restore order, to put the team back together and then move the work forward."
Murati left OpenAI in late 2024 and is now advocating a different vision for AI development, focusing on what she calls "interaction models."