OpenAI board crisis: The board fired its CEO and lost

Five days in November and the narrow claim

On Friday 17 November 2023, the board of OpenAI, Inc., the nonprofit that governed every part of the company behind ChatGPT, announced that Sam Altman would leave as chief executive and give up his board seat. Its announcement of a leadership transition gave one reason. After what it called a deliberative review, the board had concluded that Altman was "not consistently candid in his communications with the board," which hindered its ability to carry out its responsibilities, and it no longer had confidence in him. Mira Murati, the chief technology officer, became interim chief executive. Greg Brockman was to lose the board chair but stay on as president; within hours he quit.

Five days later the outcome had reversed. Late on 21 November, Pacific time, OpenAI agreed to bring Altman back, and on 29 November it confirmed his return under a new initial board of Bret Taylor as chair, the former Treasury secretary Larry Summers, and Adam D'Angelo, the only director who had voted for the ouster and kept his seat, as NPR's explainer noted.

The narrow, documented claim is this: a nonprofit board used the authority that OpenAI's structure gave it to remove the chief executive of the company behind ChatGPT, and within a week employees and the company's largest investor, Microsoft, had forced the decision to be undone and three of the four directors who made it to leave. Everything else said about the episode, including why it happened and what it proved, is interpretation, and much of the evidence surfaced only later, in a review summary, interviews and a lawsuit. This report keeps those layers apart: what the parties said at the time, what the 2024 review found, what later sworn testimony added, and what the governance changes that followed can and cannot be attributed to.

A board built to answer to humanity, not investors

The power the board used was designed in. When OpenAI created a capped-profit subsidiary in 2019 to raise capital, it kept the nonprofit's board as the governing body for everything. The company's own description of its structure, as archived the day before the firing, set out five principles. The nonprofit fully controlled the for-profit through a manager entity it owned. Directors owed their duties to the mission, so the principal beneficiary was humanity rather than investors. The board was majority independent, and independent directors held no equity. Returns to investors and employees, Microsoft included, were capped. And the board alone would decide when OpenAI had reached artificial general intelligence, a technology carved out of Microsoft's licenses. The page stated that Microsoft had no board seat and no control, despite an investment commitment that NPR put at 13 billion dollars.

The bylaws made the mechanism blunt. As Paresh Dave reported in Wired, the January 2016 bylaws let directors elect and remove one another and allowed a majority to act by written consent without notice or a meeting. During 2023 three directors left, Reid Hoffman, Shivon Zilis and Will Hurd, shrinking the board to six. Three were executives: Altman, Brockman and the chief scientist Ilya Sutskever. Three were outsiders: D'Angelo, the entrepreneur Tasha McCauley, and Helen Toner of Georgetown's Center for Security and Emerging Technology. The structure page still called the board majority independent, but on Wired's count it was split evenly. Sutskever and the three outsiders formed a majority able to remove both the chief executive and the chair without either of them in the room.

The design was meant as a check on what OpenAI's founding had framed as the central danger: commercial pressure overriding a mission to make advanced AI safe and broadly beneficial. The management scholar Alnoor Ebrahim, writing in The Conversation, described it as hybrid governance, in which a nonprofit board's first task is preventing mission drift. He also observed that half the directors were paid executives of the organization they oversaw, which nonprofit practice treats as a conflict of interest.

What the board said, and what it left unsaid

The board added little to its announcement in public. The Verge reported that employees learned of the decision when it was published, that OpenAI's communications team declined to comment beyond the blog post, and that Brockman wrote that Sutskever had told him and Altman of their removal from the board. Microsoft issued a statement from Satya Nadella the same day, saying it had a long-term agreement with full access to what it needed and remained committed to Murati and her team.

Frictions existed before the firing, though the board cited none. In October 2023 Toner had co-written a study for her center, Decoding Intentions, on how governments and companies signal restraint. One case study set OpenAI's release of ChatGPT and GPT-4 against Anthropic's choice to hold back its chatbot Claude until a comparable product was public, reading that delay as a signal against the "frantic corner-cutting" that ChatGPT's release appeared to spur. Both Toner ( The Register) and Sutskever ( his deposition) later described the paper as a source of tension. At the time, though, the only stated ground was candor, and the gap filled with unverified accounts, including an anonymous letter from self-described former employees urging the board to hold firm against profit-driven pressure, as NPR reported.

The weekend the leverage shifted

Over the weekend the pressure ran one way. The Verge reported on 18 November that the board was discussing Altman's return, and on Sunday he posted a photograph of himself wearing a visitor's badge at OpenAI's headquarters, according to Wired's account of the staff revolt. The talks failed. Late on 19 November the board told staff that Altman would not return and replaced Murati with a second interim chief executive, Emmett Shear, formerly of Twitch. Hours later Nadella announced that Altman and Brockman, with colleagues, would join Microsoft to lead a new advanced AI research team, while Microsoft stayed committed to its OpenAI partnership and looked forward to working with Shear ( Microsoft's statement).

On Monday 20 November the staff answered. An open letter told the board that its process had jeopardized the company's work, that it had never supplied written evidence for its allegations despite repeated requests, and that it had told the leadership team that allowing the company to be destroyed "would be consistent with the mission." The signatories said they might resign and join the new Microsoft unit, which they said had positions for all of them, unless every director resigned, two new lead independent directors were appointed (they suggested Taylor and Will Hurd) and Altman and Brockman were reinstated. By that evening Wired counted 738 signatures among about 770 employees; NPR later gave 702. One signer was Sutskever, who had posted that he deeply regretted his part in the board's actions.

This was the decisive turn. The board's legal authority was intact, but the company it governed consisted of people and computing capacity, and both were offering to move to Microsoft. Employees shared in the capped-profit subsidiary's returns, as the 2023 structure page explained, so the board's decision put their own stake at risk along with the company. Later testimony suggests the board weighed unusual options. In a 2025 deposition, Sutskever said that over that weekend the directors held a call with Anthropic's leaders, Dario and Daniela Amodei among them, about a proposal that Anthropic merge with OpenAI and take over its leadership. He opposed it, the other directors were more receptive, and the talks were extremely brief, ending over practical obstacles Anthropic raised ( Sutskever's deposition). He did not know which side made the first approach, and no other participant's account of the call is public.

Return, review and a rebuilt board

The reversal came late on 21 November, the date OpenAI's later review gives for the decision to rehire Altman and Brockman. Nadella called the change in the board a first essential step toward more stable governance ( Microsoft). On 29 November OpenAI published messages from Altman and Taylor ( OpenAI). Murati returned as chief technology officer and Brockman as president, Sutskever left the board, and Microsoft would hold a non-voting observer seat. Altman thanked Toner and McCauley for their effort and said the company had not lost a single employee or customer. Taylor promised a qualified, diverse board, an independent committee to review the events, and stronger governance. On 8 December that committee, Taylor and Summers, retained the law firm WilmerHale.

The review's findings were released only as a summary, on 8 March 2024 ( OpenAI's summary of the WilmerHale review). WilmerHale reviewed more than 30,000 documents and conducted dozens of interviews. It found that trust between Altman and the old board had broken down, and that the board's November statement fairly described its decision and reasons. In WilmerHale's account the decision had nothing to do with product safety or security, how fast OpenAI was developing its technology, its finances, or what it had told investors, customers and business partners. The board had moved fast, without warning key stakeholders, completing a full inquiry or letting Altman answer its concerns. Firing him was within its wide discretion, yet nothing in his conduct required it. The board then expressed full confidence in Altman and Brockman, Altman rejoined it, and three directors were added: Sue Desmond-Hellmann, Nicole Seligman and Fidji Simo. OpenAI also adopted corporate governance guidelines, a stronger conflict-of-interest policy, a whistleblower hotline and new board committees. The full report has not been published.

The accounts that came later

The departed directors' later accounts went further than the board's 2023 statement. In May 2024, on the TED AI Show, Toner said the board learned of ChatGPT's launch on Twitter, named Altman's ownership of the OpenAI Startup Fund as another example of what it was not told, and said that on multiple occasions he gave the board inaccurate information about the company's few formal safety processes. She also alleged that after her paper appeared, Altman lied to other directors to try to push her off the board. Taylor replied that OpenAI was disappointed she continued to revisit these issues, and repeated that the decision had not been about safety or finances ( The Register).

Sworn testimony followed in Elon Musk's lawsuit against Altman and OpenAI over the company's move away from a pure nonprofit. In a deposition on 1 October 2025, Sutskever confirmed that he had sent the independent directors a 52-page memo at their request, most likely D'Angelo's, whose first page said Altman showed a "consistent pattern of lying, undermining his execs, and pitting his execs against one another" ( deposition transcript). He testified that he thought termination was the right action, that he had been considering it for at least a year, that he used a disappearing email for fear of leaks, and that he wrote a similar memo about Brockman. Under questioning he acknowledged that most of the screenshots came from Murati and that he had not checked some second-hand claims with the people concerned. He also recalled Toner saying, in a meeting with executives, something to the effect that the company's destruction could be consistent with the mission, and said he had at one point supported asking her to leave the board after her paper appeared.

At trial in May 2026, jurors heard Toner describe a pattern of behavior concerning Altman's honesty and candor, McCauley allege repeated crises under his leadership, and Murati say he told different people opposite things, while Sutskever confirmed his memo from the stand. Nadella testified that no one ever gave him a specific reason for the firing and called the board's handling "amateur city" ( The Guardian). Altman and OpenAI denied Musk's allegations. On 18 May the jury returned an advisory verdict that Musk's claims were barred by the statute of limitations, and the judge accepted it ( MIT Technology Review). The verdict settled nothing about the 2023 dispute itself.

From a nonprofit veto to a committee with a halt button

The governance that replaced the old board moved oversight from one blunt power, firing the chief executive, toward standing procedures. In May 2024 the board formed a Safety and Security Committee chaired by Taylor that included Altman. In September 2024 it became an independent board oversight committee chaired by Zico Kolter of Carnegie Mellon, without Altman, with authority alongside the full board to delay a model release until safety concerns were addressed; it had already reviewed the safety assessment for the o1 release.

The deeper question, whether a nonprofit should control the company at all, was then reopened. According to the Delaware Department of Justice, OpenAI's original plan would have ended the nonprofit's oversight of the business. After engagement with Attorney General Kathy Jennings and California's Rob Bonta, OpenAI announced in May 2025 that the nonprofit would retain control. The recapitalization closed on 28 October 2025 ( OpenAI). The nonprofit, renamed the OpenAI Foundation, controls OpenAI Group PBC, a public benefit corporation, and holds 26 percent of its equity, valued then at about 130 billion dollars, while Microsoft holds roughly 27 percent ( OpenAI's structure page). Delaware's statement of no objection carried conditions. The Foundation appoints and can remove every PBC director. On safety and security questions, PBC directors may consider only the mission, not stockholders' financial interests. The Safety and Security Committee stays with the nonprofit, with power to require mitigations "up to and including halting the release of models or AI systems," and its chair sits only on the Foundation board ( Delaware Department of Justice).

Whether this answers the 2023 failure is disputed. Ebrahim argued in a 2026 essay that because all but one director sat on both boards, the Foundation largely oversaw itself, and noted that OpenAI's mission statement in its latest tax filing no longer included the word safely. At the Musk trial, Musk's counsel pointed out that seven of the Foundation's eight directors were on the for-profit board, and the Northwestern law professor Jill Horwitz told MIT Technology Review that it was unclear how the nonprofit could exercise its duties over the company.

The honest limits

Several gaps limit any confident reading. The board never published detailed reasons in 2023, the WilmerHale report survives only as a company-issued summary, and the most detailed accounts come from participants with interests of their own, often given years later in adversarial litigation. Sutskever's memo has not been released, and he conceded it rested partly on unverified second-hand claims. The record also pulls in two directions on motive. The review says the decision was not about product safety or the pace of development, while Toner frames the loss of trust partly through inaccurate reporting on safety processes. Both can be true, a dispute about candor in which safety reporting was one subject, but the public record does not resolve it.

Causal claims about what followed are provisional too. The 2025 recapitalization had drivers of its own: OpenAI said its mission now required hundreds of billions of dollars, its 6.6 billion dollar funding round in late 2024 was conditioned on converting to a more conventional company, as Ebrahim notes, and two attorneys general shaped the final terms. It is defensible to say that the crisis discredited the old board's way of exercising control and preceded the safety committee that Delaware later entrenched, but not that it caused the new structure. Nor does the episode show what a board with more experience, advance notice to stakeholders and a documented case would have achieved.

How to place the board crisis in AI history

The defensible judgment is institutional, not technical. As far as the public record shows, the board crisis was the first time the governance machinery a frontier lab built to restrain itself was turned against its own chief executive, and it showed that formal authority was not the same as power. The board could remove Altman by majority vote, but it could not keep him removed once nearly all employees, whose financial stake depended on the company, and the investor that was also its compute partner ( OpenAI's 2023 structure page) stood behind him. Toner drew the lesson that a company regulating itself is not enough ( The Register), and Ebrahim argued in 2023 for regulation with teeth ( The Conversation). Later public rules, among them California SB 53, moved part of that question from private boards to statute. Whether OpenAI's 2025 structure, with its safety committee answerable in part to state regulators, would let a future board prevail in a similar test remains open, and the episode's lasting influence should be treated as provisional.

References

• OpenAI (2023). OpenAI announces leadership transition.
• OpenAI (2023). Sam Altman returns as CEO, OpenAI has a new initial board.
• OpenAI (2024). Review completed & Altman, Brockman to continue to lead OpenAI.
• OpenAI, archived by the Internet Archive (2023). Our structure, as published on 16 November 2023.
• OpenAI (2024). OpenAI Board Forms Safety and Security Committee.
• OpenAI (2024). An update on our safety & security practices.
• OpenAI (2025). Evolving OpenAI's structure.
• Bret Taylor, OpenAI (2025). Built to benefit everyone.
• OpenAI (2025). Our structure.
• Satya Nadella, Microsoft (2023). A statement from Microsoft Chairman and CEO Satya Nadella.
• Jay Peters and Alex Heath, The Verge (2023). Sam Altman fired as CEO of OpenAI.
• Will Knight and Steven Levy, Wired (2023). OpenAI Staff Threaten to Quit Unless Board Resigns.
• Paresh Dave, Wired (2023). How OpenAI's Bizarre Structure Gave 4 People the Power to Fire Sam Altman.
• Bobby Allyn, NPR (2023). Why CEO Sam Altman's firing at ChatGPT maker OpenAI was bound to happen.
• Alnoor Ebrahim, The Conversation (2023). OpenAI is a nonprofit-corporate hybrid: a management expert explains how this model works, and how it fueled the tumult around CEO Sam Altman's short-lived ouster.
• Andrew Imbrie, Owen Daniels and Helen Toner, Center for Security and Emerging Technology (2023). Decoding Intentions: Artificial Intelligence and Costly Signals.
• Richard Speed, The Register (2024). Ex-OpenAI board member accuses Sam Altman of 'outright' lies.
• U.S. District Court for the Northern District of California, via CourtListener (2025). Videotaped Deposition of Ilya Sutskever, Musk v. Altman, No. 4:24-cv-04722-YGR.
• Nick Robins-Early, The Guardian (2026). 'A consistent pattern of lying': Musk v OpenAI trial exposes what insiders think of Sam Altman.
• Michelle Kim, MIT Technology Review (2026). Musk v. Altman week 3: Musk and Altman traded blows over each other's credibility. Now the jury will pick a side.
• Delaware Department of Justice (2025). AG Jennings completes review of OpenAI recapitalization.
• Alnoor Ebrahim, The Conversation (2026). OpenAI has deleted the word 'safely' from its mission, and its new structure is a test for whether AI serves society or shareholders.