IA 360
Current Affairs

Sam Altman returns to OpenAI: five days that reveal who really holds power in a company

Altman’s return did not erase board power; it exposed the difference between legal authority and the bargaining power of staff, investors and partners.

4 min read AI-generated Leer en español
Sam Altman returns to OpenAI: five days that reveal who really holds power in a company

On Tuesday, November 21, 2023, US time, OpenAI announced an agreement in principle for Sam Altman to return as chief executive. The official confirmation described an “initial” board consisting of Bret Taylor as chair, Larry Summers and Adam D’Angelo. That word matters: the group was not presented as final, and D’Angelo was not new; he had served on the board that removed Altman four days earlier.

The sequence looks like a story of victory and defeat, but it offers a more useful lesson in corporate governance. Boards, management, employees, investors and commercial partners exercise different powers. The board had formal authority to dismiss the CEO. Employees and Microsoft did not need that vote to make the decision costly. Legal authority and bargaining power met for several days and then collided.

Friday: what the document said and withheld

On November 17, OpenAI’s original statement announced that Altman would leave the CEO role and board. It said a review had concluded that he was not “consistently candid” in communications, hindering the board’s responsibilities, and that the board had lost confidence. Mira Murati became interim CEO. No specific conduct, date, financial transaction or safety problem was identified.

That gap limited what could be stated on November 21. The official sentence supported reporting a breakdown of trust; it did not support filling the cause with rumors. A corporate statement can be primary evidence of what a board decided and alleged without independently proving that its characterization is correct. The original source has authority over the act, not every dispute around it.

The same statement explained the structure: the board of OpenAI, Inc., a 501(c)(3) nonprofit, was the overall governing body for OpenAI’s activities. It recalled that the organization created a for-profit subsidiary in 2019 to raise capital while preserving the nonprofit’s mission, governance and oversight. This helps explain why a large investor could not simply vote to restore the CEO like a conventional controlling shareholder.

The weekend: titles are not interchangeable

After the dismissal, Murati temporarily led the company and Greg Brockman stepped down as board chair; the initial announcement said he would remain at the company, although he then announced his departure. On Sunday, the board selected Emmett Shear as another interim CEO. Two temporary chiefs in four days did not mean either received an ordinary mandate. An interim executive preserves operations through a transition; the board retains appointment and oversight.

Other words also need separation. A board chair organizes the body that governs and supervises; a CEO runs the business; a cofounder has historical influence but not necessarily special legal rights; an investor contributes capital under contracts; and an employee may leave subject to employment terms. Headlines compress these roles into “leaders,” but their levers differ.

OpenAI’s Charter says its mission is to ensure artificial general intelligence benefits all humanity and that its primary fiduciary duty is to humanity. That is extraordinary language compared with a conventional corporation. Yet a mission principle does not specify the procedure for firing a CEO, informing partners or retaining employees. Governance also requires bylaws, contracts and board rules, not all of which were publicly available.

Monday: Microsoft creates a credible alternative

On November 20, Satya Nadella said in a public statement that Microsoft looked forward to Altman and Brockman leading a new advanced AI research team. The message also reaffirmed commitment to OpenAI’s partnership and team. This was not an order to the board; it was an alternative route that could attract people while preserving Microsoft’s access to talent and capability.

That day, a letter signed by more than 700 employees demanded the board’s resignation and threatened a move to the Microsoft organization. The document circulated through journalists and social platforms rather than as a stable OpenAI corporate page. The count should therefore be attributed to the letter and people who verified it, not presented as a flawless official census.

The threat was powerful because a research lab’s central asset is a team coordinating hard-to-replace knowledge. But “more than 700 signatures” does not prove that every person would have resigned, that Microsoft would have hired all of them or that transfers would have been frictionless. A collective letter is evidence of intention and cohesion at that moment, not an observed exodus.

Tuesday: agreement in principle, not complete restoration

The return announcement was deliberately narrow. It said agreement in principle, initial board and collaboration to work out details. It confirmed Altman’s position and three names; it did not publish the agreement, a complete division of powers or a final explanation for the dismissal. “The board surrendered” and “Microsoft took control” are interpretations of the sequence, not contents of the document.

Bret Taylor, former Twitter board chair and Salesforce co-CEO, became chair. Former US Treasury Secretary Larry Summers joined. Adam D’Angelo remained. The change was substantial but not total, and “initial” suggested later expansion. Describing it as a board built for safety, Microsoft or Altman would have required information the announcement did not provide.

The calendar deserves precision too. The dismissal occurred Friday the 17th and the return agreement arrived Tuesday the 21st in the United States, four days later; Europeans saw the post on Wednesday the 22nd. “Five days” can describe five inclusive calendar dates, but not five complete days. Dating each event prevents a narrative formula from replacing the clock.

How to map power

For another corporate crisis, build a matrix. Put the board, CEO, employees, shareholders, creditors, customers, cloud provider and regulators in rows. In columns, record formal authority, the resource each controls, available exit and the cost of exercising it. OpenAI’s board controlled appointment; Microsoft controlled capital, cloud and an employment alternative; employees controlled continuity of knowledge; Altman controlled leadership and relationships; the nonprofit held the formal mission.

Then separate capacity from legal right. Microsoft could offer jobs, but the available announcement gave it no reinstatement vote. Employees could threaten to leave but could not directly dismiss directors. The board could appoint a CEO but could not compel the workforce to trust that person. A party with a credible alternative can exert influence beyond its formal representation without making the legal structure disappear.

The final check is documentary. Link every announcement, record its time, preserve its verbs and note what is absent. “Agreed,” “looks forward,” “interim” and “in principle” are not synonyms for “executed,” “hired,” “permanent” and “final.” During a fast crisis, those distinctions separate facts from forecasts that may change within hours.

The company continued; the questions did not disappear

As of November 21, OpenAI had prevented Altman’s announced departure and created a new board core. It had not published the specific cause of the breakdown, an independent report or a final governance architecture. There was also no evidence for attributing the crisis to AI safety, a financial transaction or one individual. The rigorous answer to those gaps was “not established.”

The transferable skill is reading corporate power without confusing it with an organization chart. First identify who can legally make a decision. Then identify who controls the resources required to make it work. Check the credible alternatives available to each party. Finally, return to the documents and separate provisional from final. Altman’s return shows that a board can win a vote and lose the room to sustain it; it does not show that the board was irrelevant.

This article was produced with artificial intelligence under human editorial oversight.

Share this article

This website uses cookies to improve the browsing experience. Cookie policy.

↑↓ navigate ↵ open esc close