OpenAI came within 48 hours of ceasing to exist in November 2023.

That is what the co-founder of OpenAI testified under oath Monday in a federal courtroom in Oakland.

Over a single weekend after the board fired Sam Altman on November 17, Anthropic reached out with a proposal to merge with OpenAI and take over its leadership under Dario Amodei. The proposal lasted 48 hours. Microsoft offered to absorb the entire OpenAI staff. 738 of OpenAI’s 770 employees signed a letter threatening to quit. The board’s authority collapsed in four days.

The witness called his ouster vote a “Hail Mary.”

His personal OpenAI stake, disclosed for the first time on the witness stand: $7 billion. Up from $5 billion six months ago.

This was Ilya Sutskever, OpenAI’s former chief scientist, testifying in Musk v. OpenAI before Judge Yvonne Gonzalez Rogers.

For at least a year before the board vote, Sutskever had been considering Altman’s removal. He prepared, at the request of the independent directors, a 52-page memo. The opening, per his deposition: Altman “exhibits a consistent pattern of lying, undermining his execs, and pitting his execs against one another.” Most of the evidence was secondhand screenshots from then-CTO Mira Murati. Sutskever delivered the memo to the independent directors only, through a disappearing email link.

The Anthropic outreach landed within hours of the firing. Helen Toner relayed the proposal. Dario Amodei and Daniela Amodei joined the call. Anthropic “expressed their excitement,” per Sutskever. He was “very unhappy” about it. “I really did not want OpenAI to merge with Anthropic.”

Practical obstacles, most likely the tangled investor agreements on both sides, killed the talks.

Four days later, Sutskever voted with the rest of the board to reinstate Altman. He said he “didn’t want it to be destroyed.”

Microsoft CEO Satya Nadella testified earlier the same day. He called the OpenAI board’s handling of the firing “amateur city.”

Helen Toner and Natasha McCauley voted with Sutskever to fire Altman. They testified earlier that Altman was untrustworthy, withheld information, and sometimes lied. Both resigned after his reinstatement. Three of the four directors who voted to fire him are no longer on the board.

One specific lie disclosed in their testimony: that Altman falsely told colleagues OpenAI’s in-house legal team had determined that GPT-4 Turbo did not require a safety review.

The man who tried to remove Altman over alleged threats to OpenAI’s safety mission has gained roughly $2 billion in six months from the for-profit pivot he opposed.

Altman is testifying today. Closing arguments Thursday. Jury deliberations begin Monday.

This is the trial in which a $150 billion damages claim sits on top of the question of whether OpenAI’s founding mission was a charitable trust.

The witness who almost ended OpenAI’s independent existence revealed under oath that the CEO testifying today survived a 48-hour window only because 738 employees signed a letter and Microsoft drew up a contingency plan most people in the room did not know existed.

This is not corporate drama.

This is the empirical physics of frontier AI governance under stress.

The board lost.