Sam Altman and the Long Road Back to OpenAI’s Boardroom

There is no shortage of men in this business who can give a good interview. I have known a few of them, and most were better company than the companies they ran. There is exactly one who was fired by his own board on a Friday and reinstalled as chief executive by the following Wednesday, with most of the company threatening to walk out if he was not. I have seen men voted out of offices, clubs and riverboat partnerships, and I have never once seen the vote reversed by the employees. That is a new thing under the sun, and I take my hat off to it, though I am not sure I approve.

Whatever else that week accomplished, it settled a practical question that had been hanging over the company since its founding, and settled it without any help from lawyers. In a contest between the chief executive and the non-profit board, the staff and the investors were going to side with the chief executive. Everything else was commentary.

A building put up for a different weather

The company was designed as a capped-profit concern controlled by a non-profit board whose stated purpose was to ensure that artificial general intelligence benefited humanity rather than shareholders. That was a genuinely novel arrangement, and for several years it worked well enough that nobody outside the building examined it closely. Arrangements of that kind generally work well right up until the afternoon they do not.

The crisis exposed the tension built into the design, the way high water exposes what the carpenters did. A board with a mission mandate and no commercial stake can act on that mandate — and when this one did, the commercial side of the company pushed back hard enough to reverse the decision outright. I do not say the board was wrong on the merits. I say only that a fence holds until somebody leans on it.

What was rebuilt, and what was merely repainted

Since then the board has been reconstituted, the governance documents have been revised, and the company has moved further toward conventional corporate norms. The mission language remains, and I have no reason to doubt its sincerity. The commercial pressure has, if anything, only thickened.

For those of us watching from the street, the lesson is less about personalities than about incentives — the duller reading and almost always the correct one. Any organisation that asks a small group of volunteers to hold a veto over a company worth hundreds of billions of dollars is running a structural experiment. That experiment now has data, and the data is not ambiguous.

Three things still written in pencil

• How much independent authority does the board actually retain over deployment decisions?
• What becomes of the mission commitment if the company ever needs capital on terms the board dislikes?
• Can the non-profit and the for-profit arms disagree in public without destabilising either one?

Altman’s public posture since that week has been notably institutional: more talk of safety processes, more appearances alongside regulators, less of the improvisational manner that defined his earlier years. Whether that reflects a real change of heart or merely a change of register is the thing his critics and his admirers still argue about, and I have noticed that the two camps rarely disagree about the facts, only about what the facts mean. I have sat on both sides of that argument, usually at once.

The growth that will not wait

The governance question would be a good deal simpler if the company were not growing at the rate it is. Every quarter brings more users, more enterprise customers, larger capital commitments, and more employees whose rent depends on the organisation continuing to function. Nothing in this world makes men more agreeable to the arrangement in front of them than a mortgage.

That growth is itself a constraint on oversight. A board that takes a decision the staff dislikes faces a different arithmetic when the staff number in the thousands and the company’s software sits inside other companies’ products. A veto is a cheap thing to hold over a small shop. It gets expensive fast when the shop has become the road.

What other boards learned from it

Other laboratories drew a lesson from the episode, and not the flattering one. The outcome was plain enough: a mission-oriented board which tried to exercise its authority was defeated by commercial interests. Several organisations subsequently wrote governance documents giving their boards less room to act unilaterally, which is a curious thing to do after watching somebody get overruled — and entirely rational.

Whether that is a retreat from accountability or a plain recognition of how companies actually work, nobody in this trade has settled yet. I have my suspicions and will keep them. What is clear is that the arrangement as originally designed is unlikely to be reproduced by anybody who watched it fail — not by repeal, but by nobody volunteering to build the second one.