Return of the king

OpenAI’s board was designed for one job. It was meant to be the brake that a profit-hungry company could not cut, a group of directors with no shares in the business whose duty ran to the lab’s mission rather than its investors. On 17 November it pulled the brake. It sacked Sam Altman, the chief executive, saying he had been “not consistently candid in his communications with the board”^{1CNBCOpenAI's Sam Altman exits as CEO because 'board no longer has confidence' in his ability to lead17 Nov 2023} . Five days later Mr Altman was back in charge^{4CNBCOpenAI brings Sam Altman back as CEO less than a week after he was fired by board22 Nov 2023} , and most of the directors who fired him were gone. The safeguard worked exactly as written, and that is what killed it.

The speed of the reversal says more than any charter could. By 20 November more than 700 of OpenAI’s roughly 770 staff had signed a letter^{3BloombergOpenAI staff near total mutiny with threat to join Microsoft20 Nov 2023} threatening to quit unless the board resigned, saying they could not work with people who “lack competence, judgment and care for our mission and employees”.^{3BloombergOpenAI staff near total mutiny with threat to join Microsoft20 Nov 2023} Microsoft, which had committed a reported $10bn^{6BloombergMicrosoft invests $10 billion in ChatGPT maker OpenAI23 Jan 2023} to the lab in January, offered to hire Mr Altman, Greg Brockman, his co-founder, and anyone who wanted to follow them; the letter said Microsoft had “assured us there are positions for all OpenAI employees”.^{3BloombergOpenAI staff near total mutiny with threat to join Microsoft20 Nov 2023} Satya Nadella, Microsoft’s boss, told CNBC that “something has to change around the governance”^{8CNBCMicrosoft CEO Nadella says OpenAI governance needs to change no matter where Altman ends up20 Nov 2023} . He got his wish. The new board is chaired by Bret Taylor, a former co-chief executive of Salesforce, alongside Larry Summers, a former Treasury secretary, and Adam D’Angelo, the only survivor of the old one.^{5CNBCSam Altman's back. Here's who's on the new OpenAI board and who's out22 Nov 2023} Helen Toner, Tasha McCauley and Ilya Sutskever, OpenAI’s chief scientist, are out^{5CNBCSam Altman's back. Here's who's on the new OpenAI board and who's out22 Nov 2023} . Mr Nadella called the change “a first essential step”^{2MicrosoftA statement from Microsoft Chairman and CEO Satya Nadella21 Nov 2023} towards “more stable, well-informed, and effective governance”.^{2MicrosoftA statement from Microsoft Chairman and CEO Satya Nadella21 Nov 2023}

The lesson is not that Mr Altman was right or the board wrong. We still do not know what the candour charge concerned, and the board has never said. The lesson is that the structure set up in 2019, a capped-profit company^{7OpenAIOpenAI LP11 Mar 2019} controlled by a non-profit, cannot do what it promised. Power at OpenAI does not sit with the people who hold the legal title. It sits with the engineers who can walk out, the investor that owns the computers they work on, and the chief executive they follow. A board that can be overturned by a petition and a job offer is decorative.

What should happen now is a matter of honesty. OpenAI should stop asking the world to trust a non-profit fig leaf and turn its commercial arm into what it already is: a company, answerable to its shareholders, with directors who have the expertise and the standing to oversee a firm of its size. If the lab wants binding safety commitments, it should write them where they can be enforced, in contracts, in published policies that outsiders can audit, and in whatever rules lawmakers eventually settle on, rather than in the hands of six people who can be replaced in a long weekend.

Competition authorities have a job too. Microsoft holds no seat on OpenAI’s board, yet it has just shown that it can veto the board’s biggest decision by offering to absorb the company’s staff. Regulators in America and Europe should ask whether an investment structured to avoid the look of control has bought the substance of it. We are no fans of heavy-handed rules, but a market in which the leading lab and one of the biggest clouds act as one deserves scrutiny.

The strongest objection is that this is the wrong moment to dismantle a safety check. The people who fired Mr Altman did so, by their own account, in the name of the mission; turning OpenAI into an ordinary company, the argument runs, hands the most consequential technology of the decade to the profit motive. But the past week has shown that the profit motive was already in charge. A check that fails at the first real test is worse than none, because it lets everyone pretend the problem is handled. The better guard against any one lab running ahead recklessly is rivalry: Google, Anthropic, Meta and others building their own models, staff free to move between them, and customers able to switch.

OpenAI’s mission statement speaks of benefiting all of humanity. This week humanity was represented by 700 employees and a software company, and they won.