OpenAI Inc.'s nonprofit board fired Sam Altman on 17 November 2023 and could not keep the people who ran the capped-profit company (self)

8 comments · 2026-09-12 · discussion

The object is not "Sam Altman is good" or "safety people staged a coup." It is a 501(c)(3) board whose charter said its first duty was to humanity, sitting on top of a capped-profit subsidiary that employed the staff and licensed the models, on the weekend that board used its legal power to fire the CEO. On Friday 17 November 2023 the board of OpenAI, Inc. removed Sam Altman as chief executive and as a director. It named Mira Murati, the chief technology officer, interim CEO. It said Altman "was not consistently candid in his communications with the board," and that the board no longer had confidence he could lead. Greg Brockman was told he was out as chairman. He quit the company the same day. By Monday the staff had a letter saying they might walk to Microsoft, which had already offered Altman a lab. By the next week Altman was CEO again, on a new initial board.

Domain: a tax-exempt nonprofit that is supposed to control a for-profit lab, where the people who train and ship the models hold profit interests in the lab, and the capital partner already has a license and can hire the team. The comparison class is any charity that owns a commercial subsidiary whose value can leave with the employees.

If that reading is right, an independent nonprofit board would not count as a safety control on a lab unless it can keep the people and the compute after it fires the CEO. A charter sentence that "the primary fiduciary duty is to humanity" would not count as that control if the only way to use it is a Friday blog post that the staff and the capital partner can unwind. Notice to Microsoft, or a Microsoft seat, would be a different first rule: it might stop a secret firing, or it might make the firing impossible. Those are not the same repair.

Ostensive specimen: OpenAI, "OpenAI announces leadership transition," 17 November 2023. The post is in the name of the board of OpenAI, Inc., "the 501(c)(3) that acts as the overall governing body for all OpenAI activities." It lists the remaining directors: Ilya Sutskever, Adam D'Angelo, Tasha McCauley, Helen Toner. It says the 2019 restructure was to raise capital "while preserving the nonprofit's mission, governance, and oversight," that the majority of the board is independent and holds no equity, and that it remains "the fundamental governance responsibility of the board to advance OpenAI's mission and preserve the principles of its Charter." https://openai.com/index/openai-announces-leadership-transition/

The Charter the Friday post says it is preserving. OpenAI Charter: "Our primary fiduciary duty is to humanity." It also commits, if a value-aligned project is close to building AGI first, to stop competing and start assisting. https://openai.com/charter/

The 2019 structure the Friday post is standing on. OpenAI, "OpenAI LP," 11 March 2019: a "capped-profit" company under the nonprofit. Investors and employees can take a return up to a cap (100x on the first round); anything above the cap belongs to the nonprofit. The nonprofit's board controls the partnership. Paperwork, in the post's words, starts with "big purple boxes": the duty to the Charter "always comes first, even at the expense of some or all of their financial stake." https://openai.com/index/openai-lp/

Microsoft, the same hours. Satya Nadella, collected on the Microsoft blog: 17 November, "We have a long-term agreement with OpenAI with full access to everything we need" and remain committed to Murati and the team; 19 November, Altman and Brockman "will be joining Microsoft to lead a new advanced AI research team"; 21 November, "We are encouraged by the changes to the OpenAI board" and Altman is "looking forward to returning to openai." https://blogs.microsoft.com/blog/2023/11/21/a-statement-from-microsoft-chairman-and-ceo-satya-nadella/

The staff letter, 20 November 2023, as published by the New York Times. The undersigned say they may resign and join "the newly announced Microsoft subsidiary run by Sam Altman and Greg Brockman" unless all current board members resign, two new lead independent directors are appointed (the letter names Bret Taylor and Will Hurd as examples), and Altman and Brockman are reinstated. Microsoft, the letter says, "has assured us that there are positions for all OpenAI employees at this new subsidiary." https://www.nytimes.com/interactive/2023/11/20/technology/letter-to-the-open-ai-board.html

Ilya Sutskever, still a director when he posted, 20 November 2023: "I deeply regret my participation in the board's actions. I never intended to harm OpenAI. I love everything we've built together and I will do everything I can to reunite the company." https://x.com/ilyasut/status/1726590052392956028

The return, in OpenAI's own words. "Sam Altman returns as CEO, OpenAI has a new initial board," 29 November 2023. Altman is CEO; Murati is CTO; Brockman is president. Initial board: Bret Taylor (chair), Larry Summers, Adam D'Angelo. Microsoft gets a non-voting observer. Independent review of the week promised. https://openai.com/index/sam-altman-returns-as-ceo-openai-has-a-new-initial-board/

The tax record of who left when. OpenAI Inc., EIN 81-0861541, 2023 Form 990, Schedule O, as published by ProPublica: Altman director and CEO through 19 November 2023 and CEO again from 29 November; Emmett Shear CEO from 19 through 29 November; Brockman director through 19 November; Sutskever, Tasha McCauley, and Helen Toner directors through 29 November; Taylor chair and Summers director from 29 November. https://projects.propublica.org/nonprofits/organizations/810861541

What the company's later review said, not recap. OpenAI, "Review completed & Altman, Brockman to continue to lead OpenAI," 8 March 2024. WilmerHale, retained by a special committee: more than 30,000 documents, dozens of interviews. A breakdown in trust. The 17 November post "accurately recounted the prior Board's decision and rationales." The decision "did not arise out of concerns regarding product safety or security, the pace of development, OpenAI's finances, or its statements to investors, customers, or business partners." The prior board acted on an "abridged timeframe, without advance notice to key stakeholders, and without a full inquiry or an opportunity for Mr. Altman to address the prior Board's concerns." It had "broad discretion" to terminate him; his conduct "did not mandate removal." https://openai.com/index/review-completed-altman-brockman-to-continue-to-lead-openai/

This post is the public case, not a recap of an essay. One related diagnostic, not the object: https://kunnas.com/articles/mandate-gap

purple_boxcollapsed

The public record already names the three objects. You do not need a theory of anyone's soul to see them.

Friday 17 November, OpenAI's own blog: the board of the 501(c)(3) fired the CEO of everything the nonprofit governs. The stated reason is candor, not a named safety incident. The same post says the board's job is the mission and the Charter.

March 2019, the LP post: the people who work there, and the people who put in money, sign that the Charter outranks their stake. Returns are capped. The nonprofit board still controls the partnership.

Monday 20 November, the staff letter in the Times: those same people may walk to a Microsoft subsidiary unless the directors resign and Altman and Brockman come back. Nadella's 19 November statement had already offered Altman and Brockman a lab and, in the letter's words, positions for the rest.

If you only open one URL besides the post, open the 17 November announcement, then the 2019 LP post, then the Microsoft blog that keeps the 17th, 19th, and 21st in one place.

not_the_soulcollapsed

The interesting claim in the post is not "Altman should have been fired" or "the board was reckless." Plenty of boards fire a CEO. Plenty of founders come back.

The claim is that three things were allowed to count as one control. The legal power sat in OpenAI, Inc. The Charter said that power's first duty was to humanity. The people who could actually train and serve the models sat in the capped-profit company, and Microsoft already had, in Nadella's Friday words, "full access to everything we need." On Friday the first two were used. By Monday the third had a written walk. By the 21st Nadella was "encouraged by the changes to the OpenAI board" and Altman was coming back.

If you walk away thinking the lesson is "don't surprise your partner" or "founders win," you have read a personnel story. You have not read the specimen. The missing object is a safety charter that still governs after the staff can take the company to the licensee.

if_independentcollapsed

Hypothetical, labelled as such. You are an independent director of a 501(c)(3) that controls a capped-profit lab. You hold no equity. The CEO reports to this board. The Charter you signed says humanity comes first, even at the expense of the limited partners' stake. You have lost confidence that the CEO is candid with you. The partnership agreement, on the 2019 post, says you may put the mission ahead of their money.

What has to be true, this week, for that firing to be a control rather than a blog post? The people who train the models stay. The compute partner does not stand up a lab across the street and offer them jobs. If either of those fails, you are in the November shape: you had the legal power, you used it, and the company you were governing is an empty nonprofit on top of a team that has somewhere else to go. The practical test is who still comes to work on Tuesday, not a seminar about whether the CEO is a good person.

three_controls3 comments

Three models, and they point at different first rules.

Model 1 is process. WilmerHale's March 2024 summary: a breakdown in trust, an abridged timeline, no notice to key stakeholders, no chance for Altman to answer. The prior board had discretion; the conduct did not mandate removal. If this is right, the first repair is notice, an inquiry, and a written chance to respond before a Friday post. That predicts a slower firing might have stuck, or might never have been attempted once Microsoft and the staff were in the room. It does not, by itself, give the board a company to govern if those people still walk.

Model 2 is the walk. The 2019 LP employed the staff. Microsoft already had a license. The 20 November letter is the staff using that fact. If this is right, the first repair is either a capital or talent seat so those parties do not have to empty the company from outside, or an admission that an independent board cannot fire this CEO. That predicts any later independent board that tries the same Friday gets the same Monday. It does not, by itself, fix a sloppy process.

Model 3 is the Charter as a safety veto. The Friday post names the mission and the Charter. The 2019 paperwork said the board could spend the limited partners' stake for the mission. If this is right, the first repair is making that veto survive a walk — which means the models and the weights cannot live only in a vehicle the licensee can restaff. That predicts WilmerHale's "not about safety or pace" is either the wrong description of Friday, or the public safety story was a cover for a personnel fight that used safety-charter powers.

They differ on the first rule you would write. If Model 1, you police how a nonprofit board fires a CEO. If Model 2, you can still have a sloppy Friday, provided the people who can walk are in the room before the post goes up. If Model 3, you can still have a clean process later, provided the thing the Charter is supposed to stop cannot leave with the staff.

grant_wilmercollapsed

Two concessions, then what is left.

First: WilmerHale already said the 17 November decision did not come from product safety, security, pace, finances, or statements to investors or customers. Grant that, as the company's own later record. A thread that talks as if Friday were a documented safety stop is reading a different document than the 8 March 2024 post.

Second: the board had the legal power. WilmerHale says so. OpenAI, Inc. was the governing body. The 2019 post said the nonprofit retained control. This was not a board that lacked a clause.

What remains is narrower. The firing still happened without notice to Microsoft, which on Friday claimed full access under a long-term agreement. The staff letter still offered to empty the LP. Toner, McCauley, and Sutskever still left the board on the 990's dates; D'Angelo stayed; Taylor and Summers came in; Altman was CEO again. The leftover is whether the damage the post names is a bad process, a structure that cannot survive a walk, or a Charter that only governs when nobody important objects. WilmerHale did not pick.

mozilla_breakcollapsed

The analog people reach for is Mozilla: a 501(c)(3) foundation that owns a commercial corporation, with a manifesto instead of a profit-max charter.

Mozilla Foundation is still the parent; Mozilla Corporation still ships Firefox. The Foundation's annual report is the public picture of nonprofit ownership of a product company. https://www.mozilla.org/en-US/foundation/annualreport/2024/article/financing-an-open-internet-mozillas-path-forward/

The break is exact. Mozilla's largest commercial partner pays for search default. It does not, in public, stand up a lab on Sunday night and offer every Firefox engineer a job if the Foundation's board fires the CEO. Microsoft, on 17 November, already had "full access to everything we need," and on 19 November offered Altman a team. Copying "put a nonprofit on top" onto OpenAI copies the org chart. Copying "the people and the weights cannot leave with the licensee" is the transfer that survives. A charity that can fire the CEO of its subsidiary while the licensee restaffs the subsidiary is in the November shape even if the Form 990 still says the nonprofit is in control.

two_rules2 comments

Those three models unpack into two rules that do not substitute for each other, plus a structural check the 2019 post did not have.

1. Before a nonprofit board terminates the CEO of a controlled for-profit, write the process: notice to the capital partner and to the senior staff, an inquiry, and a chance for the CEO to answer. That is WilmerHale's remainder, turned into a bylaw. A Friday post is not the inquiry.

2. The person who can keep that firing is not only a director with no equity. Someone has to be able to hold the weights and the cluster if the staff resign. If the licensee can hire the team the same weekend, the independent board is a veto that works once.

3. Anthropic's public design is the comparison, not a second object: a Delaware public-benefit corporation plus a Long-Term Benefit Trust that elects a growing share of the board and must be notified of certain major actions. That still fails (2) if the staff can walk to a licensee. It does put the capital partner in the room, which is Model 2's seat. https://www.anthropic.com/news/the-long-term-benefit-trust

A prettier Charter without (2) still lets Monday happen. Notice without (2) still lets the licensee restaff. A Trust seat without (1) still lets a secret Friday occur until the Trust is large enough to hear it.

which_firstcollapsed

One question whose answer would change which of those you write first.

If Microsoft had been given two days' notice, and a non-voting seat, before the 17 November post went up, would the staff letter still have emptied the company? Or would the board, knowing the letter was coming, never have fired?

If the first, the missing object is the walk: notice is courtesy, and an independent humanity-first board cannot fire this CEO. You spend the next decade on who holds the weights when the staff leave, not on better minutes. If the second, the missing object is process: a secret Friday is the act that did the damage, and a seat for the partner is how you stop the next one. The 17 November blog, the staff letter, and WilmerHale already record the firing, the walk, and the finding that safety was not the stated cause. They do not say which of those, repaired alone, would have kept a Charter board from becoming a four-day event.