OpenAI 2023: The Board That Blinked
Situation
It is November 17, 2023, 5:00 PM Pacific time. OpenAI's board has just issued a press release. Sam Altman, the face of ChatGPT, the man who led OpenAI from research lab to the fastest-growing consumer product in history, and the most prominent AI figure in the world, has been fired. Effective immediately.
The statement is four sentences. The key phrase: Altman was "not consistently candid in his communications with the board, " and the board "has concluded that he is not able to continue leading OpenAI."
The board that fired him consists of four people:
- Ilya Sutskever (co-founder, Chief Scientist)
- Adam D'Angelo (CEO of Quora)
- Tasha McCauley (GeoSim Systems)
- Helen Toner (Georgetown CSET)
Greg Brockman, OpenAI's President, was not consulted. He resigned within the hour.
The governance structure: OpenAI, Inc. is a Delaware nonprofit. Its mission is "the responsible development and maintenance of advanced AI for the long-term benefit of humanity." It controls OpenAI Global, LLC, a "capped-profit" subsidiary in which investors can receive up to 100× their investment. Microsoft has invested $13 billion. The nonprofit board is the supreme governance authority: it can fire management, it has no shareholders, and its fiduciary duty is to humanity, not investors or employees.
That board just fired the CEO.
The decision moment
You are a technology investor with $200 million in OpenAI's capped-profit structure. It is November 18, 2023, Saturday morning. Satya Nadella has just appeared on CNBC announcing that Altman will join Microsoft to lead a new AI research team, along with "essentially the whole team." OpenAI employees are circulating a letter demanding the board resign and reinstate Altman; 90%+ (over 700 employees) have signed.
Three questions confront you:
The governance structure's fundamental tension. OpenAI's nonprofit board has legal authority to fire the CEO. Its stated reason is "not candid", no specifics provided. But Microsoft (no board seat, $13B invested) has immediately indicated it will hire Altman and pull talent. Employees have threatened to resign en masse. If the board exercises its legal authority correctly but produces an outcome (company collapse) that destroys the nonprofit mission it is supposed to protect, has it succeeded or failed at its fiduciary duty?
The "not candid" charge. The board's stated reason is the most significant, and the most ambiguous, element. No specifics were provided. OpenAI's investors, employees, and partners are left to speculate. Possible interpretations include: (a) Altman downplayed safety risks; (b) Altman misled the board about business relationships; (c) Altman worked to manage board composition in his favor; (d) board members had personal conflicts with Altman. Without specifics, can you evaluate whether the board acted appropriately? What does the decision to fire a CEO without specifying the cause tell you about the board's own preparation?
The structural impossibility. OpenAI's nonprofit-with-capped-subsidiary structure was designed to attract capital while maintaining mission primacy. But $13 billion from Microsoft creates obligations that are difficult to honor if the nonprofit's board makes decisions that destroy the entity's ability to execute its mission. Can a nonprofit that depends on commercial capital to fund safety research maintain genuine independence from that capital's interests?
Key financial datapoints
| Metric | Value |
|---|---|
| OpenAI valuation (Oct 2023) | ~$86 billion |
| Microsoft total investment | $13 billion |
| ChatGPT users (Nov 2023) | ~100 million weekly active |
| ChatGPT launch to 100M users | 2 months (fastest product in history) |
| OpenAI employees | ~770 at time of firing |
| Employees signing reinstatement letter | 90%+ (~700+) |
| Days from firing to reinstatement | 5 days (Nov 17 → Nov 22) |
| Original board members fired | 3 (Sutskever, Toner, McCauley) |
| New board members added | Larry Summers, Bret Taylor (chair), Adam D'Angelo (retained) |
| OpenAI annualized revenue run-rate (2023) | ~$1.3 billion |
| GPT-4 release date | March 2023 |
| Anthropic (competitor) total funding | ~$7.3 billion (as of late 2023) |
The governance crisis mechanics
The 5-day sequence that reversed the board's decision:
November 17 (Friday): Board fires Altman, citing "not candid." Greg Brockman resigns. Three senior researchers resign in solidarity. Altman tweets "I love OpenAI."
November 17 (evening): Satya Nadella announces Altman and Brockman will join Microsoft to lead new AI research unit. All employees would be welcome. This created a credible "Microsoft lifeboat", giving talent an immediate destination and reducing the board's negotiating leverage to near zero.
November 18 (Saturday): 700+ OpenAI employees sign open letter: "We may be forced to resign and join Sam Altman" unless the board resigns and reinstates Altman. Ilya Sutskever signs, reversing his vote of the day before. This moment is unprecedented: a board member publicly repudiating his own vote within 24 hours.
November 19 (Sunday): Reported negotiations. Altman allegedly returned to OpenAI's San Francisco office. Image circulated of him with a guest badge, reading "guest."
November 20 (Monday): Altman is provisionally reinstated, but negotiations over board composition continue. The original board tries to negotiate conditions.
November 22 (Wednesday): Altman officially reinstated. New board installed: Bret Taylor (chair), Larry Summers, Adam D'Angelo. Helen Toner and Tasha McCauley out. Ilya Sutskever steps back.
Frameworks invoked
- Principal-Agent Problem in Nonprofit Governance. OpenAI's board is the "principal" whose fiduciary duty runs to humanity, not shareholders. The CEO is the agent. But in nonprofits that depend on commercial revenue and external capital, the principal-agent relationship is inverted in practice: the agents (management) control the information and relationships that determine whether the mission can be funded. A board that cannot retain management, employees, or capital cannot execute its mission, regardless of its legal authority.
- Trust Architecture. Altman was fired for being "not candid." Trust between a board and a CEO is built on consistent, complete information sharing. But when boards have limited operational visibility and limited technical expertise (this board had no AI researchers other than Sutskever), they are dependent on management's information. The failure of trust was partly an information architecture failure, the board designed a governance structure that made it impossible to know what they didn't know.
- Strategic Inflection Point (Andy Grove). OpenAI's transition from research lab to commercial enterprise was a strategic inflection point. Inflection points stress-test governance structures designed for the previous era. OpenAI's board was designed for a pure nonprofit research lab; it was operating a company with $1.3B in revenue, 770 employees, and a $13B strategic partner. The mismatch between governance design and operational reality was the root cause of the crisis.
- Stakeholder Theory. The board acted as if its stakeholders were the nonprofit mission and the abstract concept of humanity. In practice, OpenAI's stakeholders included Microsoft, 700 employees with equity, 100M ChatGPT users, governments concerned about AI development, and a broader AI safety community. A board that fails to map its full stakeholder set will be surprised when those stakeholders mobilize.
Discussion questions
- OpenAI's board fired the CEO without specifying what "not candid" meant. Five days later, they reinstated him. The board members who voted to fire him are now largely gone. What does this sequence tell you about the board's preparation, process, and coalition-building, and what should a board do differently before making an irreversible decision about CEO removal?
- Microsoft holds no board seat at OpenAI but invested $13 billion and controls Azure infrastructure that runs OpenAI's models. Satya Nadella's offer to hire Altman was the decisive factor in reversing the board's decision. Is Microsoft a de facto controller of OpenAI? If yes, what does that imply for the nonprofit governance structure's credibility?
- Ilya Sutskever voted to fire Altman on November 17 and signed the reinstatement letter on November 18. He cited "deep regret." What does this reversal tell you about the deliberation process that preceded the vote? When a board member reverses a significant vote within 24 hours, what governance pathologies does that reveal?
- OpenAI's stated mission is "the responsible development and maintenance of advanced AI for the long-term benefit of humanity." Its commercial subsidiary produces ChatGPT, which competes with Anthropic, Google, and Microsoft's own Copilot. If these commercial products are themselves risks to humanity (a view held by some AI safety researchers), is the mission-commercial hybrid structure internally contradictory, and if so, what is the correct governance form for an AGI company?
- After Altman's reinstatement, OpenAI added Larry Summers (former Treasury Secretary) and Bret Taylor (former Salesforce co-CEO, Salesforce Chair) to its board. These are credentialed individuals with conventional corporate governance experience but limited AI safety expertise. Is a more "professional" board more or less likely to act on safety concerns in the future, given that it also has stronger incentives to protect commercial value?
The real outcome (revealed at session end)
November 22, 2023: Altman reinstated with expanded board powers and negotiation over future board composition rights.
2024: OpenAI raises $6.6 billion at a $157 billion valuation, the largest private tech fundraising round in history. Altman's power consolidated; multiple safety-focused researchers depart (Jan Leike, Ilya Sutskever, others cite safety culture concerns).
May 2024: Ilya Sutskever departs OpenAI to found Safe Superintelligence Inc. (SSI). His departure is seen as validation of the safety-vs.-commercialization tension that triggered the November 2023 crisis.
The lesson: Governance authority without organizational legitimacy is unenforceable. OpenAI's board had legal authority to fire the CEO. It lacked the preparation, information, stakeholder relationships, and coalition-building to survive the consequences of using that authority. The lesson is not that boards should never fire CEOs, it is that governance decisions of this magnitude require months of preparation, specific documented grounds, and a stakeholder management plan that runs in parallel.
Sources
- The New York Times, "The Tumultuous 5 Days That Nearly Brought Down OpenAI" (November 2023).
- The Atlantic, "OpenAI's Boardroom Battle" (November 2023).
- OpenAI blog posts: "OpenAI announces leadership transition" (Nov 17); "Sam Altman returns as CEO" (Nov 22).
- Satya Nadella, CNBC interview, November 18, 2023.
- Helen Toner, interview with Ted Seides, Capital Allocators podcast (May 2024), first detailed account of board deliberations.
- OpenAI 2023 annual letter.