The board fired the chief executive on a Friday.
The decision worked.
Sam Altman was no longer chief executive of OpenAI. The nonprofit board possessed the authority to remove him and used it. An interim chief executive was named. The institution announced a leadership transition.
Within days, Altman was returning.
The board was being rebuilt. An independent review had been promised. Employees, customers, technical work, leadership, and Microsoft had all entered the crisis. Microsoft had been prepared to receive the team if no settlement emerged.
That distinction is the case.
A board whose removal power works only when the chief executive agrees to be removed does not have removal power.
Neither does a board whose authority evaporates whenever employees threaten to leave, customers become nervous, investors object, or a strategic partner offers the executive another office.
Those relationships matter enormously to whether an organization survives.
They are not automatically votes.
OpenAI had deliberately placed unusual authority in a nonprofit board charged with a mission extending beyond the ordinary interests of management, investors, employees, and customers. Whatever one thinks of the November 2023 decision, an independent board cannot mean independent until independence becomes disruptive.
Confidential information may also constrain what directors can say publicly. Delay can deepen an injury. A board that has concluded it cannot trust an executive may sometimes have to act before it can assemble a process satisfying everybody who depends on the executive.
So grant the removal power.
OpenAI’s Special Committee later said the prior board possessed broad discretion to fire Altman.
Then it described how the decision had been made.
According to the published summary of the later WilmerHale review, the rupture involved a breakdown in trust, an abridged time frame, no full inquiry, no advance notice to key stakeholders, and no opportunity for Altman to answer the concerns. The summary said the board had not anticipated how severely the decision would destabilize the organization.
It also said the firing did not arise from concerns about product safety or security, the pace of development, OpenAI’s finances, or Altman’s statements to investors, customers, or business partners.
The full WilmerHale report was never published.
That limitation belongs in the case. These are the reconstructed institution’s public findings about its own rupture after Altman had returned. They are evidence. They are not an outside tribunal delivering the final truth about everything the directors knew or feared.
The narrower fact is enough.
The board could issue the decision. It had not established that it could carry what happened after the decision.
Legal authority moves through a material institution.
Someone has to lead the teams on Monday.
Someone has to know which technical work cannot pause.
Someone has to retain the people carrying that work.
Customers need confidence that the service will continue. Security obligations persist. Contracts persist. Compute arrangements persist. Research programs, releases, internal disagreements, financing, and institutional memory do not reorganize themselves because a board resolution changed a name.
The messages announcing Altman’s return named much of that operating field: leadership, employees, customers, Microsoft, technical work, and confidence in continued operation.
None acquired a legal veto by appearing in the crisis.
Their presence showed the limits of the board’s legal instrument.
Formal authority can remove a person while lacking the capacity to carry the institution after the person is gone.
That gap is easy to call founder worship once the founder is famous.
It is harder to dismiss when the work began before there was an established institution capable of judging it.
OpenAI began in a field where many of the relevant comparisons did not yet exist.
The 2015 announcement described a nonprofit artificial-intelligence research company with a founding and research team and large financial commitments entering technical territory whose important paths were still uncertain.
Somebody had to assemble the laboratory before later results could prove which decisions had been wise.
Researchers had to be found and persuaded to work together.
Compute had to be financed and scheduled.
Architecture, data, evaluation, security, release strategy, hiring, timing, and acceptable failure had to be chosen while evidence remained incomplete.
Novel work often begins with an asymmetry.
One person or a small team may know the prototypes, failures, users, technical stack, compromises, dangers, and timing with an exactness no outside committee can reproduce.
Voting on every uncertain move can turn direction into negotiated blur.
Disclosure can destroy a bargaining position or expose a vulnerability.
An incumbent institution may be excellent at judging everything already legible to it and structurally unable to receive the project that changes the comparison.
A field that refuses to recognize unequal contribution becomes incapable of recognizing creation.
That unusual contact can justify formative authority: concentrated command during the period when integration, secrecy, speed, and direction are conditions of making the thing at all.
Formative authority can be substantial.
Office.
Ownership.
Budget control.
Confidentiality.
Protected surplus.
A higher burden on someone proposing to interrupt the work.
It can last years.
The founder does not become illegitimate because other people eventually become important.
The problem begins when evidence about the project becomes title over the field surrounding it.
A successful prototype proves something about the prototype.
A functioning organization proves more.
Repeated success under difficult conditions can establish extraordinary judgment.
None of these facts freezes the original relationship.
Other people learn.
The organization changes.
Dependencies deepen.
Successors acquire knowledge.
The project becomes an employer, purchaser, platform, infrastructure user, data holder, standards maker, security target, and supplier to institutions whose lives extend far beyond the founding laboratory.
The founder’s contact changes too.
Knowing the model architecture may justify enormous weight over an architectural decision.
It does not create equivalent knowledge of a contractor’s body, the water consumed by a data center, the rights of an author, the dependency of a school, the security judgment of a government, or the life of a worker whose future now depends on the platform.
Those relations touch the project.
They do not become departments of the founder’s mind.
Local success establishes local contact; it never establishes field-wide title.
Founder title begins when demonstrated contact with the created project becomes presumptive jurisdiction over the wider people, institutions, infrastructures, and successor possibilities the project has come to reach.
The promotion happens gently.
The project expands.
Because it survived under founding judgment, continued survival becomes evidence that founding judgment should govern every new relation.
Correction starts looking like ignorance.
Refusal looks like betrayal.
Succession looks like an attack on the thing itself.
The frontier closes behind the person who once escaped through it.
Refusing founder title does not settle the OpenAI case.
It creates another temptation.
Give the board the final answer.
That fails too.
A board can hold records without holding operating judgment.
It can possess legal independence and still misunderstand technical dependence, tacit knowledge, organizational trust, or the consequences of a sudden transfer.
Its power may be necessary precisely because insiders are capable of protecting themselves.
Its distance may also make a destructive intervention look clean on paper.
OpenAI’s later review captured that tension unusually clearly. The prior board had broad authority to remove Altman. According to the institution’s later account, it used that authority through a process too compressed to establish a survivable transition.
An institution cannot solve this problem with better paperwork alone.
A succession plan saying the board may remove the chief executive is incomplete if removal also means losing the people, access, relationships, knowledge, records, technical command, and trust required to operate the next morning.
Correction has to carry capacity.
That means building the conditions under which unusual founding authority can eventually narrow without the work collapsing around it.
Knowledge has to survive outside one person’s memory.
Other leaders need real operating contact before the crisis.
Records need to preserve reasons rather than only conclusions.
Protected reporting has to reach somewhere other than the person being challenged.
Independent reviewers need enough technical competence to understand the dispute.
Emergency removal may sometimes precede notice, while ordinary disputed removal needs evidence, conflict review, inquiry, and a chance to answer.
Most importantly, the successor needs the ability to act.
Documentation cannot replace judgment. A binder cannot acquire tacit knowledge on command. A board resolution cannot keep a research program alive.
Succession has to be built while the incumbent is still present enough to help make succession possible.
That is uncomfortable.
The constitution has to distrust both conveniences.
The firing of Sam Altman does not establish that founders should rule their institutions indefinitely.
It does not establish that boards should defer whenever their decisions threaten organizational stability.
It exposes a harder requirement.
Removal is not succession.
A founder can deserve extraordinary authority without owning the institution’s future.
A board can possess genuine removal power without possessing the practical capacity to survive using it.
An institution becomes governable when the work can outlive the person and correction can outlive the office performing it.
That takes more than a succession clause.
It takes people, records, money, infrastructure, trust, technical competence, independent evidence, operating authority, and a real path through which one carrier can lose power without taking the function down with it.
The founder can lose the crown. The board does not inherit it.
And refusing both crowns decides nothing about where the chips, power, money, work, access, and losses should go next.
They still have to be allocated.