It’s a hemorrhage. OpenAI has seen its founders leave one after the other for several months. In August, John Schulman announced his departure for Anthropic, a major rival of Sam Altman’s company. This researcher had nevertheless worked a lot to secure ChatGPT, verifying that its operation was “aligned” with human values. At the same time, OpenAI president Greg Brockman announced that he was taking a sabbatical for the rest of the year. “I am taking a break for the first time since the creation of OpenAI […] I’ve put OpenAI ahead of everything, including my marriage, for the past nine years. The work we do is important to me, but so are other parts of my life,” he writes on X. Brockman is one of Sam Altman’s major supporters – he had also been abruptly dismissed from his position last November, before the board of directors backpedaled in the face of the anger that this “coup” had provoked. On X, he insists on the temporary nature of his leave. But the announcement comes at a bad time, after a series of departures.
Last May, another key figure in the company had already packed his bags: Ilya Sutskever, OpenAI’s chief scientist. The latter had supported the “coup” against Altman at the end of 2023 before changing his position and calling for his return, apologizing profusely. He left to set up Safe Superintelligence, a new company clearly very focused on the security of generative AI.
Andrej Karpathy, for his part, did not last long. This researcher had already left OpenAI once in 2017, in order to take a position at Tesla. But he returned home in 2023. In February 2024, however, he left again in order to set up an AI start-up applied to education, Eureka Labs.
There are definitely not many of OpenAI’s founders left today. Of course, several had left the adventure between 2016 and 2018, at a time when OpenAI was not yet as successful as it is today. But this recent wave of departures has drastically reduced the number of those still on board. Of the eleven who participated in the creation of OpenAI, there are now only… two left. Sam Altman, at the helm of the company, as well as researcher Wojciech Zaremba. There have also been other notable departures, outside the circle of co-founders, in particular that of Jan Leike, one of OpenAI’s security managers. A surprising brain drain for a company that was supposed to be at the height of its glory. It reveals two things.
Tensions within OpenAI are revealed
The first is that OpenAI’s approach to securing its products is far from unanimous. When he left, Jan Leike did not mince his words, assuring that for several months, his team had been “struggling to be heard” and to obtain the computing power necessary to carry out the work he deemed necessary. “Building machines that are smarter than humans is an inherently risky undertaking. OpenAI assumes an immense responsibility on behalf of all humanity. But developing flashy products has taken precedence over our security culture in recent years,” he said in a vehement post about X. Jan Leike has also joined Dario and Daniela Amodei, two former OpenAI employees who set up Anthropic, a direct competitor of the latter, which is supposed to be more invested in AI security.
These series of departures also show that Altman is caught in his own game. The head of OpenAI did not just show off the prowess – very real – of his great language models. He dangled before the world the imminent arrival of super-powerful AI, much more intelligent than humans. A questionable strategy that helped create a prodigious craze around generative AI. And triggered an intense war for brains. AI talents are being poached with big checks. And see investors afraid of missing the AI train blindly offering them generous funding to set up their start-ups. It’s not easy to retain employees under these conditions.
In the coming months, the hype around AI startups should die down. It is becoming increasingly clear that only a fraction of them have a real chance of ever breaking through the profitability barrier. Investors will not write blank checks to young companies forever. And a concentration will occur. In a way, it has already begun, as shown by the deals that Amazon, Microsoft and Google have respectively made with Adept, Inflection AI and Character AI.
OpenAI must, however, remain on its guard. In recent years, the Microsoft steamroller has smoothed its path, providing it with considerable resources (computing power, funding, etc.). In total, the tech giant has invested $13 billion in the AI nugget in exchange for 49% of its shares. Since the melodrama of last November, Satya Nadella has however understood that he should not put all his eggs in one basket. Microsoft has since invested in the French Mistral AI as well as in the Emirati start-up G42. The group is also developing its own AI (the mini-Phi models but also MAI-1, a much larger model according to The Information). In August, in a statement to the SEC, Microsoft for the first time designated OpenAI as a “competitor.” The honeymoon is over.
.