OpenAI's nonprofit kill switch fired Sam Altman, then collapsed in 72 hours under employee and investor pressure.
A few weeks earlier Ilya Sutskever sent three fellow board members a 70-page memo whose first line read: Sam exhibits a consistent pattern of lying, undermining his execs, and pitting his execs against one another. Its screenshots were photographed off personal phones, and many came from Mira Murati, who had described a toxic leadership pattern going back years.
The statement said Mr. Altman's departure followed a deliberative review process by the board, which concluded that he was not consistently candid in his communications with the board, hindering its ability to exercise its responsibilities. The same announcement named chief technology officer Mira Murati as interim CEO. No specific incident was ever given to staff.
At a December 2022 board meeting he said three controversial GPT-4 changes had all been approved by the Joint Safety Review Board; Helen Toner asked for documentation and only one had been approved. Microsoft tested unreleased GPT-4 in India with no safety approval, which the board learned by accident in a hallway conversation.
Altman's 2019 fix was a nonprofit board of directors owning and controlling a for-profit subsidiary, so the for-profit could take investment, pay startup equity and build data centers while answering to the mission. Two guardrails: investor profits capped at 100x, with everything beyond flowing back to the nonprofit, and the board's power to fire the CEO.
By Tuesday morning more than 700 of OpenAI's 770 employees had signed the letter demanding Altman and Brockman be reinstated. With over 90% of the company threatening to walk, the board knew it had lost. A deal was reached: Sutskever, Toner and McCauley out, D'Angelo staying, and at 10pm Pacific Altman's return was announced.
OpenAI tried to restructure in December 2024 so the nonprofit would essentially lose control; after months of pressure it backed down, sort of. Under the new new structure the nonprofit is on top in name, a safety committee could in theory halt model releases, and it owns about a quarter of the equity. The profit caps are gone.
He had told the board and the world he was an independent director with no financial stake in the company. In fact the OpenAI Startup Fund, $175 million of venture capital deployed in OpenAI's name, was personally owned by him. It now seems this was mostly to make paperwork easier, but clarifying it took months.
Brett Taylor, CEO of Salesforce and someone McCauley had previously worried would be too deferential to Sam, and Larry Summers, former Treasury Secretary and president of Harvard, agreed to join. Adam D'Angelo stayed and is still on the board to this day. Brockman and Altman did not get their board seats back right away.
In just a few days OpenAI was due to close a deal letting employees sell their own shares at the $86 billion valuation, meaning millions of dollars for many of them, and some had already put down deposits on houses. Asked about it at the all-hands, COO Brad Lightcap said only: the tender, we're going to see.
For most of his life he has seen things in AI before other people see them. In 2012 he and Geoffrey Hinton debuted a deep neural network that demolished the image-labelling world record, where the best programs were wrong about 25% of the time. In 2015 he turned down a $6 million a year counter-offer from Google to co-found OpenAI.
“AI will probably, like most likely sort of lead to the end of the world.”
“that could in fact be consistent with the mission.”
“No one person should be trusted here.”
“Feel the AGI.”
“I don't think Sam is the guy who should have his finger on the button.”
“You could parachute him into an island full of cannibals, come back in five years, and he'd be the king.”
“Sam Altman Fired and Rehired.” https://www.youtube.com/watch?v=_eYTkvZqbnQ. Transcript summary by WhipScribe, https://whipscribe.com/transcript-pages/sam-altman-fired-and-rehired.html.