Sutskever deposition details 52-page memo behind Altman ouster

The roadster refund was a spark. The deposition shows the fire.

Musk’s “you stole a nonprofit” jab made headlines; Ilya Sutskever’s sworn testimony shows what actually mattered. The text of a court-filed deposition transcript paints a more complicated story of how OpenAI unraveled and then re-coalesced in November 2023.

The November chaos, clarified

Sutskever told lawyers he had waited “at least a year” for board dynamics that would allow Sam Altman’s removal. When independent directors Helen Toner, Tasha McCauley, and Adam D’Angelo finally aligned, they moved in secret, fearing Altman would “pull out all the stops” if he learned of the plan. He would, Sutskever testified, “find a way to make [documents] disappear.”

Within a day of the firing, the board held a call with Anthropic leaders Dario and Daniela Amodei about a potential merger in which Anthropic would take control of OpenAI. Sutskever said he was “very unhappy” about it and recalled Toner as the “most supportive.” The talks fizzled almost immediately on “practical obstacles.” That brief window matters. It underscores how fragile OpenAI looked—and how rivals moved fast when it wobbled.

The memo that lit the fuse

Months before the ouster, Sutskever compiled a 52-page memo for the independent directors. Its opening claim was blunt: Altman showed “a consistent pattern of lying, undermining his execs, and pitting his execs against one another.” Sutskever said the memo included screenshots he obtained—“most or all” from CTO Mira Murati—and that he sent it via a disappearing email link to prevent leaks. He did not share it with Altman.

The document also recounted hearsay. Sutskever wrote that, according to Murati, COO Brad Lightcap said Altman had been pushed out at Y Combinator for similar behavior, and that Greg Brockman had been “essentially fired” from Stripe. Under oath, he said he didn’t verify either claim with Lightcap or Brockman. That caveat matters. The memo galvanized a board, but parts of it rested on second-hand assertions.

What the employees did next

Sutskever expected staff to be calm or even supportive after the dismissal. He was wrong. Within days, hundreds of OpenAI employees threatened to quit if Altman wasn’t reinstated. Sutskever then signed the letter backing Altman’s return and later told lawyers he had misread the moment. It was a sharp lesson in where power actually sits in high-growth labs: with the people who can ship models—and the platforms they depend on.

One detail from the deposition hints at why the board’s case crumpled so fast. Sutskever confirmed he still holds OpenAI equity and that its value has increased since his departure. His counsel blocked questions about how much. He also said he hadn’t received legal bills for the deposition and wasn’t sure who was paying. The optics are stark: critics of OpenAI’s direction often remain financially intertwined with its success. That’s not unique to OpenAI, but it blurs motives.

The roadster framed against governance

Seen through this lens, the roadster refund spat reads like punctuation, not plot. Altman posted screenshots showing his years-old deposit and a failed refund request. Musk shot back that the money was returned within 24 hours and added the “you stole a nonprofit” line that ricocheted across social feeds. It was vintage Musk—simple phrasing, maximum sting. The deposition shows a messier backdrop: a chief scientist lobbying to remove his CEO, a board acting covertly, and a competitor circling within hours.

Sutskever’s testimony also exposes how brittle “mission-first” structures become under competitive pressure. He opposed the Anthropic deal, but the mere fact of the call reveals priorities. When the crown jewel of AI looks leaderless, everyone reaches for it. Mission statements don’t stop that. Money and momentum do.

Three forces decided November’s outcome

Politics over principle. The people building the core systems learned, fast, that political coalition-building matters as much as technical leadership. When Sutskever underestimated the rank-and-file reaction, the board’s case collapsed.

Capital over governance. Microsoft’s backing made the status quo hard to topple. Employees facing interrupted roadmaps and equity uncertainty chose stability. Their letter made that clear. Cash flow and compute trumped charter theory.

Competition over purity. Anthropic’s brief interest showed that even safety-branded labs move aggressively when the market opens a seam. That’s not hypocrisy. It’s the industry’s current rule set.

Sutskever now runs Safe Superintelligence, a venture-backed lab promising to “scale in peace.” OpenAI presses on. Musk taunts from the sidelines while building his own stack. The arguments about who “stole” what will continue. The incentives won the week.

Why this matters

• The fight isn’t safety vs. speed; it’s competing power centers trying to steer the same curve.
• Governance theater collapses under money, talent, and compute pressure; design for that reality, not around it.