CASE 03
The OpenAI Board Crisis
A governance structure built to intervene, discovering it could not
OpenAI was deliberately constructed so that a non-profit board with no financial stake could overrule the commercial arm if the mission required it. Investors were warned in advance that they might never see a return. In November 2023 that board used exactly the power it had been given: it removed the chief executive. Five days later he was back, most of the board was gone, and the structure that had been designed to make intervention possible had demonstrated, in public, that it was not.
Course anchor
Ch. 6 (required) + Ch. 7–18 frameworks
Case period
2015 – Mar 2024 (structure still changing)
Decision point
17–22 Nov 2023 — five days
The mirror of Case 01. The DAO was built so that nobody could intervene — and when intervention was needed, the community did it anyway by rewriting the rules. OpenAI was built so that someone could intervene — and when that someone did, the organization reassembled itself around the decision until it was reversed. Both cases ask the same question from opposite directions: is formal authority real if using it costs you the organization?
What this case is designed to teach
After analyzing this case, you should be able to:
- Separate formal authority from actual power (Ch. 11 and Ch. 17), using a case where one party held all of the first and almost none of the second.
- Analyze a governance structure on its own terms — a non-profit board controlling a capped-profit company, with a stated duty to humanity rather than to shareholders — and assess whether such a design can work.
- Reason carefully under genuine factual uncertainty. The board never fully explained itself. You will have to build an analysis that does not depend on knowing what you cannot know.
- Map stakeholders when one of them is “humanity” — a stakeholder named in the organization’s constitution, with no representative, no vote, and no voice.
- Evaluate a communication decision (Ch. 14) in which a board made a serious public accusation in eleven words and then said nothing for months.
- Assess collective employee action (Ch. 13, Ch. 16) where roughly nine in ten staff threatened to resign — and where their convictions and their financial interests pointed the same way.
Part A — The Case
Read the whole narrative before looking at the questions. Note carefully throughout which facts are established and which are not.
A.1 An unusual constitution
OpenAI was founded in 2015 as a non-profit research organization. Its stated purpose was not to make money but to ensure that artificial general intelligence — AI systems broadly more capable than humans at economically valuable work — would benefit all of humanity. Its published Charter committed it to that mission over any competing interest, including a clause undertaking to stop competing with and instead assist another project if that project were close to succeeding first.
Frontier AI research turned out to require capital and computing power on a scale no donation base could supply. In 2019 OpenAI adopted a hybrid structure that is central to this case:
- A non-profit entity, governed by a board, sits at the top and controls everything below it.
- Beneath it, a “capped-profit” commercial company does the work, raises investment, and sells products. Investor returns are capped at a multiple of the money invested; anything above the cap flows back to the non-profit.
- The board’s legal duty runs to the mission, not to investors. Investors were reportedly told in writing that they should regard their investment in the spirit of a donation and might not get their money back.
Microsoft invested at scale — roughly one billion dollars in 2019 and, according to widely reported figures, around ten billion announced in early 2023 — and supplied the computing infrastructure the research ran on. It held a substantial economic interest in the commercial entity. It did not hold a seat on the non-profit board.
Why the structure matters
This is not a technicality. Almost every organization you will study this semester is ultimately accountable to owners who want it to succeed commercially. OpenAI deliberately built the opposite: a controlling body with no financial stake, whose job was to be willing to damage the business if the mission demanded it. The design anticipated exactly the kind of conflict that arrived. Whether it survived that conflict is the question of this case.
A.2 By November 2023
The launch of ChatGPT in late 2022 turned a research laboratory into one of the fastest-growing consumer products in history. OpenAI became commercially central to Microsoft’s product strategy, its employees held equity expected to be worth a great deal in a forthcoming share sale, and the organization was hiring and shipping at high speed.
The board at the time consisted of six people: the chief executive Sam Altman; the president Greg Brockman; the chief scientist Ilya Sutskever; and three external directors — Adam D’Angelo, Tasha McCauley, and Helen Toner. A board of six, controlling an organization of roughly 770 employees and a product used by hundreds of millions of people.
A.3 Five days
On 17 November 2023 the board announced that Sam Altman was removed as chief executive. The stated reason was that he “was not consistently candid in his communications with the board,” and that the board no longer had confidence in his ability to lead. Mira Murati, the chief technology officer, was named interim chief executive. Greg Brockman was removed as chair and resigned shortly afterwards.
Two things about that announcement matter more than its content. First, it was an accusation of a serious kind — a public statement that a chief executive had not been honest with his own board — offered with no supporting detail. Second, major stakeholders learned of it with very little notice. Microsoft, the largest investor and the supplier of the computing infrastructure, was reportedly informed only moments before the announcement.
What followed was not a governance process. It was a contest of power conducted in public over a weekend. Investors pressed for reversal. Employees organised. Within days, roughly 700 of about 770 employees signed a letter stating that they would resign unless the board itself resigned and Altman was reinstated — a letter signed even by Sutskever, a board member who had participated in the removal and who publicly expressed regret. Microsoft announced that Altman and Brockman would lead a new research group inside Microsoft, and that any OpenAI employee who wished to join would be hired on equivalent terms. An interim chief executive from outside was appointed and served for a matter of days.
On 21–22 November 2023, Altman returned as chief executive. A new initial board was formed, chaired by Bret Taylor and including Larry Summers and Adam D’Angelo. Toner and McCauley left. The directors who had exercised the board’s authority no longer held it.
A.4 What was never established
This is the part of the case that requires the most discipline from you. The board never published a detailed account of its reasons. An enormous amount was written about what those reasons might have been, most of it resting on anonymous accounts of private conversations.
What can be said with more confidence is what a subsequent review reported. An external law firm was engaged to review the events, and in March 2024 the outcome was announced: the review concluded that the prior board had acted within its discretion, but that its decision did not arise from concerns about product safety or security, the pace of development, OpenAI’s finances, or its statements to investors, customers or business partners. It was described instead as arising from a breakdown in the relationship and a loss of trust. Altman rejoined the board.
The trap most reports fall into
Students arrive at this case believing it was a fight about AI safety — cautious researchers versus a commercially aggressive chief executive. It is a compelling story and it is what most commentary at the time assumed. The review that examined the question reported otherwise. If your report asserts that the removal was about safety, you are asserting something the available record does not support, and you will lose marks for it. You may discuss the safety-versus-speed tension as context — it is real and it matters — but you may not present it as the established cause.
A.5 Where the power actually was
Strip away the personalities and the case becomes a clean problem in organizational design. Six people held the entire formal authority. Everyone else held something else:
- Employees held the capability. An AI laboratory is its researchers; the assets walk out of the building every evening. Collective resignation was not a protest, it was a credible threat to dissolve the organization.
- Microsoft held the computing infrastructure, the capital, and — decisively — an open offer of employment that made the employees’ threat costless to them.
- The chief executive held relationships: with staff, with investors, with partners, and with the public. None of it was formal authority, and all of it was mobilisable within hours.
- The board held the legal right to do what it did, and nothing else. It had no independent access to the employees’ loyalty, the infrastructure, or the capital.
The board was not overruled by any legal process. No court reversed it. It exercised a power it genuinely possessed, and the organization simply re-formed itself in a shape where that power no longer existed.
A.6 Aftermath
- The board was rebuilt and expanded. New directors were added through 2024, and Microsoft took a non-voting observer seat for a period before relinquishing it. Verify the current composition before writing about it.
- Several senior safety-focused researchers left in the period that followed, some stating publicly that they believed safety work had been deprioritised. Treat individual statements as the views of the people who made them, not as findings.
- The structure itself was reopened. OpenAI subsequently pursued a significant restructuring of the relationship between the non-profit and the commercial entity. This is still developing. Check the current status before relying on any description of the structure; anything written even a year ago may be out of date.
- The governance question outlived the episode. The design OpenAI attempted — mission control over commercial operation — is being attempted again by other organizations in AI and beyond. Whether November 2023 proved the design unworkable, or merely proved that this board executed it badly, is a live argument and the most interesting question in the case.
Part B — Exhibits
You may reproduce these exhibits, but you must extend them with your own analysis. Copying them unchanged earns no analytical credit.
Exhibit 1 — Timeline
| Date | Event | Why it matters for the analysis |
|---|---|---|
| 2015 | OpenAI founded as a non-profit with a mission-first Charter. | Establishes a stated purpose beyond profit — the socioeconomic view written into a constitution (Ch. 6). |
| 2019 | Capped-profit subsidiary created under non-profit control; Microsoft invests. | The structural compromise at the heart of the case (Ch. 11). |
| Nov 2022 | ChatGPT launches; extraordinary growth follows. | Commercial stakes rise sharply, changing every incentive in the system. |
| Early 2023 | Further large Microsoft investment reported. | Dependence on a single partner for capital and compute deepens (Ch. 3 environment). |
| 17 Nov 2023 | Board removes Altman: “not consistently candid in his communications with the board.” Brockman removed as chair, resigns. | The decision itself — formal authority exercised, in eleven words, with no supporting detail (Ch. 14). |
| 17–18 Nov | Investors and partners react; Microsoft reportedly given minimal notice. | Stakeholders with no formal governance role begin to determine the outcome (Ch. 17). |
| 19–20 Nov | ~700 of ~770 employees sign a letter threatening to resign; Microsoft offers to hire all of them; an outside interim CEO is appointed. | Collective action converts into decisive power (Ch. 13, Ch. 16). The threat is credible because of Microsoft’s offer. |
| 21–22 Nov | Altman returns; new initial board (Taylor, Summers, D’Angelo); Toner and McCauley depart. | The reversal. Note that no legal authority overruled the board. |
| Mar 2024 | External review reported: board acted within its discretion; the decision did not stem from safety, security, pace, finance, or statements to partners. Altman rejoins the board. | Removes the safety narrative. The ethics analysis must proceed without it. |
| 2024 → | Board expanded; senior safety researchers depart; the non-profit/commercial structure is reopened. | Still developing. Verify before citing. |
Exhibit 2 — Formal authority versus actual power
This is the analytical core of the case. Use the five bases of power from Ch. 17 — legitimate, reward, coercive, expert, referent — and note how little correlation there is between the first column and the last.
| Party | Formal authority | Actual sources of power | Outcome |
|---|---|---|---|
| The non-profit board | Total. Legal right to appoint and remove the chief executive; control of the commercial entity. | Legitimate power only. No control of capital, compute, staff loyalty, or public narrative. | Exercised its authority successfully, then ceased to exist in its previous form within five days. |
| Employees | None. | Expert power (they are the organization’s capability) and coercive power (credible collective exit). | Decisive. The letter, not any legal mechanism, reversed the decision. |
| Microsoft | None — no board seat. | Reward power (capital, compute, an open job offer to every employee); the practical alternative venue for the whole enterprise. | Made the employees’ threat credible at no cost to them. Arguably the most consequential actor without a vote. |
| The chief executive | None once removed. | Referent power (loyalty), expert power, and an unusually wide network among investors, staff and press. | Reinstated. Later rejoined the board. |
| “Humanity” | Named in the Charter as the ultimate beneficiary. | None. No representative, no vote, no channel. | Not present in the decision at any point, in either direction. |
Exhibit 3 — Stakeholder inventory (starting point only)
| Stakeholder | What was at stake | Represented in the decision? | How the episode ended for them |
|---|---|---|---|
| Employees | Their work, their colleagues, and equity expected to be worth a great deal in a forthcoming sale. | Not at all — until they represented themselves. | Got the outcome they demanded. |
| Microsoft and other investors | Billions in capital and a central plank of product strategy. | No board seat; reportedly minimal notice. | Outcome restored; later gained an observer seat for a period. |
| The removed directors | Their fiduciary duty to the mission, and their personal reputations. | They were the decision. | Left the board; subject to sustained public criticism. |
| The chief executive | Position, reputation, and a public accusation of a lack of candour. | Not consulted in the removal. | Reinstated; later rejoined the board. |
| Users and customers | Continuity of a service many had built businesses on. | No. | Service continued; the episode revealed concentration risk. |
| Safety researchers inside the organization | Whether caution retains institutional standing. | Divided; not organised as a group. | Several departed in the following period. |
| The wider AI research community | Whether mission-controlled governance is viable anywhere. | No. | Inherited a highly visible precedent. |
| The public / “humanity” | The beneficiary named in the Charter. | No mechanism exists. | Unrepresented throughout. |
| Regulators | Oversight of a strategically significant technology. | Absent. | Attention increased sharply afterwards. |
Exhibit 4 — Three readings of the same five days
| Reading | Core claim | What follows if it is right | Strongest objection to it |
|---|---|---|---|
| The board failed Process |
The authority was real; the board simply used it incompetently — no stakeholder preparation, no succession plan, no explanation, no coalition. | Mission-control governance can work. It needs directors who can execute, and procedures decided in advance. | Even a well-executed removal would have faced the same employee and investor response. Competence may not have been the binding constraint. |
| The structure failed Design |
A board with no capital, no operating role, and no relationship with staff cannot exercise authority over an organization whose assets are its people. | This form of governance is decorative. Anyone relying on it for AI oversight is relying on nothing. | It ignores that the board did act, and that the structure survived in modified form. A single episode may not settle the design question. |
| The system worked Corrective |
A small group attempted a drastic action without explaining it; the people affected refused; the action was reversed. That is accountability functioning. | Concentrated authority should be checkable by those it governs. The episode is a feature. | The check came from parties with financial interests in one outcome. A check that only ever points one way is not a check. |
Exhibit 5 — Key figures
5 days
From removal to reinstatement
~700 / ~770
Employees signing the letter threatening to resign
6
Directors holding the entire formal authority
11 words
The public reason given for removing the chief executive
Data caution. Employee and headcount figures are as widely reported at the time and were not audited. Investment figures are reported, not disclosed in audited accounts. Attribute them, and prefer the structural facts — which are verifiable — to the numbers.
Part C — The Assignment
C.1 How this case satisfies the syllabus requirements
| Req. | Syllabus wording | How you meet it in this case |
|---|---|---|
| (1) | Select a real organization or industry case and analyze it using the frameworks covered in this course. | OpenAI is a real, documented organization with an unusual and well-published governance structure. Apply Ch. 7–18 frameworks in Questions 6–9 — at minimum structure (Ch. 11) and leadership/power (Ch. 17). |
| (2) | Include an explicit business ethics and social responsibility analysis: identify the ethical issues, map the stakeholders, assess conduct against ethical criteria and CSR expectations (Ch. 6). | Questions 2, 3, 4 and 5 are mandatory. This case is unusually rich for Ch. 6 because the organization put the socioeconomic view into its own constitution. |
| (3) | Close with a recommended course of action and the reasoning that supports it. | Question 10 — two parts: what the board should have done, and how such a structure should be designed. |
| (4) | Written report and oral presentation in English. | Part G provides terminology and presentation language support. |
C.2 First decision: which organization are you analyzing?
State your unit of analysis in the first paragraph and hold it for the whole document.
(a) The non-profit board as a governing body
The actor that made the decision. Best for fiduciary duty, Ch. 13 group decision making, and Ch. 14 communication.
(b) OpenAI as a whole organization
The full hybrid entity. Best for Ch. 11 structure and the mission-versus-commerce tension. The most common choice.
(c) Microsoft
A corporation with a vast investment and no governance rights, deciding how to act in a partner’s crisis. An unusual and rewarding choice.
(d) The employees as a collective actor
The group that decided the outcome. Best for Ch. 13 and Ch. 16 — but you must handle the equity interest honestly.
C.3 Four traps to avoid
- Asserting the removal was about AI safety. See A.4. The review that examined this reported otherwise. This is the defining sourcing error in this case.
- Reporting private conversations as fact. Much of the published narrative rests on anonymous accounts. Where you use them, attribute them as reporting and say they are unconfirmed.
- Casting a hero and a villain. Every actor here has a defensible account of their own conduct. A report that finds one party entirely right has stopped analyzing.
- Describing the current structure from memory. The non-profit/commercial arrangement has been actively restructured since these events. Verify before you describe it, and date your description.
Part D — Analysis Questions
Questions marked REQUIRED must be answered by every team. From the questions marked CHOOSE, answer at least three, including at least one from the Ch. 11 / Ch. 17 pair.
Context and unit of analysis REQUIRED
Define your unit of analysis (C.2) and describe the external environment in late 2023 — an industry with enormous capital inflows, intense competition for a small pool of researchers, essentially no binding sector-specific regulation, and extreme public attention. Then characterise the organizational culture using the Ch. 3 dimensions, and identify the specific cultural tension the structure institutionalised: a research culture with a stated mission of caution operating inside a product organization competing at maximum speed. Was this a strong culture or two weak ones?
Frameworks: Ch. 3 external environment components; environmental uncertainty; dimensions of organizational culture; subcultures.
Identify and rank the ethical issues REQUIRED
Identify at least four distinct ethical issues, each stated as a decision faced by a named actor. Candidates include: making a public accusation about a named individual’s honesty without publishing supporting evidence; removing a chief executive without notifying the principal investor or preparing the organization; signing a collective resignation letter while holding equity whose value depended on the outcome; a major corporation offering to absorb an entire partner organization’s workforce during that partner’s governance crisis; and accepting investment for a mission-first entity whose investors have every incentive to resist mission-first decisions.
Rank them using issue intensity (Exhibit 6.7). Note the analytical difficulty this case presents: the most visible harm was to a small number of powerful individuals, while the harm the organization’s Charter was actually written to prevent is diffuse, distant, and hypothetical. Explain what happens to an ethics framework built around identifiable victims when the stakeholder the organization exists to serve is an abstraction.
Frameworks: Exhibit 6.7 issue intensity; ethical dilemmas; the limits of victim-centred ethical reasoning.
Map the stakeholders REQUIRED
Extend Exhibit 3, recording stake, power, legitimacy, representation and outcome for each party. Then answer three harder questions:
- Can “humanity” be a stakeholder? The Charter names it as the ultimate beneficiary. It has no representative, no vote, and no way to be consulted. Argue whether a stakeholder with no voice is a genuine stakeholder or a rhetorical device — and if genuine, describe the mechanism that would give it one.
- Were the employees defending the mission or their equity? A share sale was anticipated and its value depended on the outcome. State plainly that both motives could operate at once, then take a position on what follows for how much moral weight the letter should carry. Avoid both cynicism and credulity.
- What did Microsoft owe? It held no governance rights and acted to protect an enormous investment — entirely rational. But its offer to hire every employee is what made the collective threat credible. Assess whether a party with no formal role acquires ethical duties when its actions determine another organization’s governance outcome.
Frameworks: Stakeholder identification and prioritization; power–legitimacy–urgency; Ch. 6 CSR expectations beyond shareholders.
Assess conduct against ethical criteria and CSR expectations REQUIRED
This question has three parts, all required.
(a) Apply the ethics tests. Take the five tests in Exhibit 6.4 and apply them to three decisions: (i) the board’s decision to remove the chief executive with an unexplained public accusation, (ii) the employees’ decision to sign the letter, and (iii) Microsoft’s decision to offer employment to the entire workforce. Present a 3×5 matrix with a verdict in each cell. The What-If-Everybody-Did-This test is the sharpest instrument here — apply it particularly to (iii), and ask what happens to every mission-driven organization if this response becomes standard whenever a board makes an unpopular decision.
(b) Classify the CSR posture. This case is unusual: OpenAI adopted the socioeconomic view formally, in its founding documents, rather than as a policy statement. Assess whether its conduct in November 2023 was consistent with that commitment, using Exhibit 6.2 to distinguish social obligation, responsiveness and responsibility. Then ask the hard question: if an organization writes the socioeconomic view into its constitution and then behaves like any other commercial organization under pressure, what does that tell us about the value of such commitments generally? Consider both the cynical and the charitable reading.
(c) Codes and charters. Ch. 6 says the effectiveness of a code of ethics depends heavily on whether management supports it and how violations are treated. OpenAI’s Charter is a public document containing real commitments. Evaluate it against Ch. 6’s criteria for an effective code, and state what would have to be true for a charter to constrain an organization when following it is expensive.
Frameworks: Exhibit 6.4 ethics tests; Exhibit 6.2; Exhibit 6.1; classical and socioeconomic views; codes of ethics and their effectiveness.
Explain the behavior REQUIRED
Use Exhibit 6.5 to explain conduct rather than judge it. Every actor here was sophisticated and none was obviously acting in bad faith.
- Structural variables. Ch. 6 identifies reward allocation as a powerful influence on behavior. Employees held equity; directors held none. Explain how that single asymmetry shaped what each group found it natural to do, without implying that anyone was bribed.
- Moral development (Exhibit 6.6). Place the board’s reasoning and the employees’ reasoning on Kohlberg’s stages and defend the placements. The board can be read as principled (acting on a duty at personal cost) or as conventional (applying its formal rules without regard to consequences). The employees can be read as Stage 3 (loyalty to a group and its leader) or as Stage 5 (rejecting an exercise of authority made without justification). Argue it.
- Where was the process? Identify which Ch. 6 mechanism — ethics training, an independent review, a protected channel for raising concerns, a pre-agreed procedure for grave decisions — would have changed the outcome, and state honestly whether the problem was the absence of a mechanism or the absence of time.
Frameworks: Exhibit 6.5 determinants; Exhibit 6.6 stages of moral development; structural variables and reward allocation; organizational culture.
Structure: can mission control a business? CHOOSE
Describe the hybrid structure using Ch. 11 concepts: chain of command, centralization, authority, responsibility, accountability. Identify precisely where authority and accountability separate — the board was accountable to a mission with no constituency capable of holding it accountable, while the parties who could hold it accountable in practice had no formal standing.
Then evaluate the design directly. What would a controlling board need, beyond legal authority, for its authority to be exercisable? Consider: independent funding, a role in operations, a relationship with employees, a pre-agreed and published procedure, staggered terms, a larger membership. For each, state what it would cost and what new problem it would create. Close with a judgement: is mission-controlled governance a workable organizational form, or a structure that works only while nobody tests it?
Frameworks: Ch. 11 organizational design, authority vs. responsibility vs. accountability, centralization; corporate governance; Ch. 9 structure follows strategy.
Power and leadership without authority CHOOSE
Using the five bases of power (Ch. 17), analyse Exhibit 2 in depth. The board held legitimate power and essentially nothing else; every other party held one or more of the remaining four. Explain why legitimate power alone proved insufficient, and identify the conditions under which it is sufficient in a conventional firm — what does a normal board have that this one did not?
Then assess leadership on both sides. Evaluate the reinstated chief executive’s mobilisation of referent and network power against the transformational/transactional distinction. Evaluate the directors: they acted at evident personal cost, which is one marker of principled leadership, and they did so without building any coalition, which is a marker of poor leadership. Can both be true? Finally, compare with Case 01, where informal expert and referent power moved a US$150 million decision in a community with no formal authority at all. What general proposition about authority survives both cases?
Frameworks: Ch. 17 bases of power, leadership styles, transformational leadership; Ch. 11 authority; comparison with Case 01.
Groups, collective action, and motivation CHOOSE
Analyse two group decisions with Ch. 13. First, the board of six: assess it against the advantages and disadvantages of group decision making, and consider whether groupthink is a plausible diagnosis — being careful, since a small group reaching an unpopular conclusion is not automatically groupthink. Second, the employee letter: roughly nine in ten staff signed within approximately 48 hours. Examine the dynamics — norms, conformity pressure, the visibility of signatures, and the cost of being among the few who did not sign.
Then apply a motivation framework from Ch. 16 to explain why people signed, distinguishing as far as the evidence allows between belief in the leader, belief in the mission, loyalty to colleagues, financial interest, and social pressure. State what evidence would be needed to weigh these, and acknowledge that you do not have it.
Frameworks: Ch. 13 group decision making, groupthink, norms and conformity; Ch. 16 motivation theories; Ch. 15 attitudes and behaviour.
Communication and crisis decision making CHOOSE
The board made a grave public accusation in eleven words and then declined to elaborate for months. Using Ch. 14, analyse that choice from both sides: the case for silence (confidentiality, legal exposure, fairness to an individual, avoiding an unwinnable public argument) and the case against (an accusation without evidence is itself a harm; stakeholders cannot evaluate a decision they cannot understand; a vacuum is filled by speculation). Then state what you would have published, and why it would have been both truthful and defensible.
Using Ch. 2, classify the removal decision: programmed or non-programmed, structured or unstructured, and made under certainty, risk or uncertainty. Identify the failures of contingency planning (Ch. 8): what foreseeable second-order consequences appear not to have been planned for, and what does this suggest about the difference between having the authority to act and being ready to act?
Frameworks: Ch. 14 communication and crisis communication; Ch. 2 decision types and conditions; Ch. 8 contingency planning.
Your recommendation REQUIRED
(a) The board’s decision. Place your team in the board meeting on 16 November 2023. You have concluded, on grounds you consider serious, that you have lost confidence in the chief executive. You do not have to decide whether that conclusion was correct — assume you hold it in good faith. Recommend what the board should do. Your recommendation must state: the decision criteria and their weights; the sequence of steps and who is informed when; what you publish and in what words; how you secure the organization’s continuity before acting; who bears the cost of your approach; the strongest argument against it; and the conditions under which you would not act at all. “Do nothing” is a permissible recommendation — but then you must say what the board’s authority is for.
(b) The structural design. Design the governance for a new organization whose stated mission may conflict with its commercial interest. Address at least: how the controlling body is composed, funded and removed; what it must publish when it acts; what procedure applies to grave decisions, agreed in advance; what rights investors and employees have and what rights they explicitly do not; how the mission’s intended beneficiaries are represented; and what prevents the structure being dissolved by the parties it constrains. For each element, name the Ch. 6 mechanism it implements and the specific failure in this case it addresses. State the cost of your design — strong mission control raises the cost of capital and may make the organization uncompetitive, which is a real objection you must answer rather than ignore.
(c) Transferability. In one paragraph: what does this case tell managers in ordinary firms about the relationship between formal authority and the ability to use it? Apply it to a situation they will actually face — a manager with the authority to make a decision that the people who must implement it do not accept.
Frameworks: Ch. 2 decision-making process; Ch. 6 codes of ethics, independent reviews, protected channels; Ch. 8 contingency planning; Ch. 11 governance design; Ch. 14 crisis communication.
Optional extensions
- Comparison with Case 01. The DAO could not intervene and did; OpenAI could intervene and could not make it hold. Write the comparative analysis and state the principle that survives both.
- Change management (Ch. 7). Treat the removal as an attempted organizational change and the employee letter as resistance. Which resistance-management techniques were available in the time the board had, and would any have worked?
- Strategy and dependence (Ch. 9). Analyse the mutual dependence between OpenAI and its principal investor as a strategic relationship. What does the episode reveal about the risks of concentration on both sides?
- Human resources (Ch. 12). Employee equity was central to this case. Examine how compensation design creates alignment with some stakeholders and misalignment with others, and what a mission-first organization should do about it.
Part E — Deliverables
Written report
- Length: 3,000–4,000 words of body text (roughly 10–14 pages), excluding cover page, exhibits, and references.
- Language: English throughout.
- Format: A4, 12 pt, 1.5 line spacing, numbered pages and numbered section headings.
- Citations: Every factual claim and every number carries a source. Consistent style (APA 7th recommended). A reference list is mandatory.
- Contribution statement: a short table naming which member led which section, signed by all members.
Required structure
- Executive summary (max. 250 words, written last)
- Case background and unit of analysis (Q1)
- Ethics and social responsibility analysis (Q2–Q5) — the largest section
- Management analysis (your three chosen questions from Q6–Q9)
- Recommendation and reasoning (Q10)
- Limitations and what would change our answer
- References · Appendices · Contribution statement
Oral presentation
- Time: 15 minutes presentation + 5 minutes Q&A (confirm in class).
- Participation: every team member speaks for a meaningful segment.
- Slides: maximum 15. Any slide that is only a paragraph of text will cost you marks.
- Language: English throughout, including Q&A.
Suggested timing
- 2 min — The governance question in one sentence, and your recommendation up front
- 2 min — The structure and the five days (assume no prior knowledge)
- 6 min — Ethics and stakeholder analysis
- 3 min — Formal authority versus actual power
- 2 min — Recommendation, costs, strongest counter-argument
Prepare for: You say the board was wrong — wrong to act, or wrong in how it acted? If the board should never use its power, why does it have any? Whose money pays for your governance design?
Sourcing and academic integrity
- Prefer primary sources: OpenAI’s own Charter and published statements, the announcements made at the time, the summary of the external review, and any statements made by the individuals concerned under their own names.
- This case has more journalism than evidence. A great deal of what circulates describes private conversations reported anonymously. Where you use such material, attribute it and label it unconfirmed.
- Do not assert the board’s reasons. They were not published. The external review reported what the decision was not about; that is the strongest statement available to you.
- Where a fact is contested or still developing — particularly the current corporate structure — say so and date your statement. Correctly flagging uncertainty earns marks.
- If you use generative AI tools, disclose what you used them for in an appendix. You remain fully responsible for every claim. Note the irony and the risk: a tool asked about this case will readily produce confident, invented detail about private board deliberations.
Part F — Grading Rubric
Total 100 points, mapped to the four syllabus requirements. The final group report is worth 30% of the course grade.
| Pts | Criterion | Excellent | Weak |
|---|---|---|---|
| 15 | Case understanding & framing Req. (1) · Q1 |
Unit of analysis declared and held; the structure described accurately and dated; established facts kept separate from reported ones; the safety narrative correctly handled. | Asserts the removal was about safety; reports anonymous accounts as fact; describes an out-of-date structure; drifts between OpenAI and Microsoft. |
| 30 | Ethics & CSR analysis Req. (2) · Q2, Q4, Q5 |
Four or more issues identified and ranked; the “humanity as stakeholder” problem confronted; the socioeconomic-view-in-a-constitution question engaged seriously; Exhibit 6.4 applied with conflicting verdicts. | Frameworks named but not applied; hero-and-villain framing; the employees’ equity interest ignored or used as a cheap accusation. |
| 15 | Stakeholder mapping Req. (2) · Q3 |
Goes beyond the exhibit; power vs. legitimacy sharply distinguished; Microsoft’s position as a decisive actor with no formal role analysed; the unrepresented beneficiary taken seriously. | Exhibit 3 reproduced; only the famous individuals discussed; no distinction between having a stake and having power. |
| 15 | Management frameworks Req. (1) · Q6–Q9 |
At least three frameworks from Ch. 7–18 applied to generate insight; the authority/power distinction used analytically rather than asserted. | Frameworks decorate the report; the bases of power listed but not applied; conclusions unchanged if the frameworks were deleted. |
| 15 | Recommendation & reasoning Req. (3) · Q10 |
Specific and sequenced; the published wording actually drafted; the cost of strong mission control acknowledged and answered; falsifying conditions given. | Vague (“better governance”); costless; assumes a board can simply be given more power without asking who would accept that. |
| 10 | English communication Req. (4) |
Clear, well-organized English; governance terminology used correctly; presentation delivered, not read; Q&A handled with composure. | Meaning obscured by language errors; terminology misused; slides read aloud; Q&A avoided. |
Note on fluency: grammatical errors that do not obstruct meaning are not penalized. Misusing a management concept is penalized, because that is a content error. In this case, confusing authority, power and accountability is a content error — the whole case turns on the distinction.
Part G — EMI Toolkit
G.1 Key terms
| Term | Meaning in this case |
|---|---|
| Board of directors | The body with legal authority to appoint and remove senior executives and to oversee the organization on behalf of whoever it is accountable to. |
| Non-profit | An organization that may earn revenue but cannot distribute profits to owners; it has no shareholders. Here, the controlling entity. |
| Capped-profit | An unusual arrangement in which investor returns are limited to a multiple of investment, with the excess flowing to the controlling non-profit. |
| Fiduciary duty | A legal duty to act in the interests of another party. The distinguishing feature here is that the duty ran to a mission rather than to shareholders. |
| Authority | The formal right, attached to a position, to make decisions and expect compliance. |
| Power | The actual capacity to influence outcomes, which may exist entirely without authority. Keep these two words apart at all times — this case is about the gap between them. |
| Accountability | Being answerable for outcomes to some party. Note the difficulty here: the board was accountable to a mission, and a mission cannot hold anyone to account. |
| Candour | Frankness and full disclosure. “Not consistently candid” was the board’s stated reason — a serious charge, deliberately unspecific. |
| Interim chief executive | A temporary appointment while a permanent successor is found. Two were appointed within five days. |
| Equity / vesting / tender offer | Employee ownership stakes, the schedule on which they are earned, and a sale event allowing employees to convert them to cash. Central to understanding the employees’ position. |
| Collective action | Coordinated action by a group that no member could take alone. The employee letter is the textbook example. |
| Governance | The system of rules and structures by which an organization is directed and controlled — as distinct from management, which is running it day to day. |
| Charter | A published statement of an organization’s purpose and binding commitments. Assess it against Ch. 6’s criteria for an effective code of ethics. |
| Observer seat | A right to attend board meetings without voting. Information without authority — a useful concept for this case. |
G.2 Useful language for the presentation
Being precise about evidence
- “The board’s reasons were never published, so we analyse the decision process rather than its merits.”
- “This was reported by journalists citing anonymous sources; we treat it as unconfirmed.”
- “The external review stated what the decision was not about. That is as far as the record goes.”
Separating authority from power
- “The board had the authority. It did not have the power.”
- “No legal mechanism reversed the decision. The organization reorganised around it.”
- “Microsoft had no vote and was arguably the decisive actor.”
Handling questions
- “We’re not claiming the removal was justified or unjustified — we don’t have the evidence. We’re assessing how it was done.”
- “Both motives could be true at once. Let me say what follows either way.”
- “That structure has since changed; our analysis is dated to November 2023.”
Avoid these
- “The board fired him over AI safety.” — not supported by the record
- “The employees only cared about money.” — unevidenced, and it discards the analysis
- “They should have communicated better.” — not a recommendation; say what should have been published, and when
Part H — Sources and Further Reading
Primary sources (start here)
- The OpenAI Charter and the organization’s own published description of its structure. Read the Charter before anything else — the whole case is about whether a document like this can bind an organization.
- OpenAI’s announcements of 17–22 November 2023, and the March 2024 statement reporting the outcome of the external review.
- Public statements made under their own names by the individuals involved, including former directors.
- Microsoft’s public statements during the episode.
Course materials
- Robbins, S. P. & Coulter, M., Management, 15th ed., Global Edition, Pearson, 2021 — Ch. 6 (Exhibits 6.1–6.8), Ch. 2, 3, 8, 11, 13, 14, 16, 17.
Secondary sources
Extensive journalism exists, and there is a growing academic literature on AI governance and hybrid organizational forms. Search your library databases for “hybrid organization” AND governance and for “AI governance” AND (board OR “corporate structure”). Use journalism for the sequence of events; use the academic literature for the structural analysis; and remember that almost nothing published about the board’s private reasoning is verified.
A note on the facts in this document. This case summarises the widely reported public record. The corporate structure has been actively restructured since these events, headcount and investment figures were reported rather than audited, and the board’s reasoning was never disclosed. Verify anything you rely on, date your description of the structure, and report uncertainty where it exists. Do not cite this case document itself as the source of a fact in your report.