OpenAI dissolves the Superalignment team after Sutskever and Leike exit

Background

Frontier AI developers employ researchers to work on alignment: the problem of making a system behave the way its designers and users intend, including in situations where the system is more capable than the people supervising it. Alignment covers a spread of technical questions — how to state a goal precisely, how to tell whether a model is pursuing it, how to keep human oversight meaningful as models become stronger. The name of OpenAI's dedicated group, the Superalignment team, pointed at the hardest case: alignment for systems more capable than the people overseeing them. Jan Leike was one of the team's co-leaders.^[1] Ilya Sutskever was a co-founder of OpenAI and its chief scientist.^[1][3]

The departures

On May 14, 2024, Sutskever announced that he was leaving OpenAI after nearly a decade at the company.^[1][3] OpenAI named its research director, Jakub Pachocki, to take over the chief scientist role.^[1][2]

Hours after Sutskever's announcement, Leike said on X that he was resigning too. His last day at OpenAI was Thursday, May 16, 2024.^[1]

Leike's public criticism

On May 17, 2024, Leike gave his reasons in public posts. Safety at OpenAI, he wrote, had "taken a backseat to shiny products". He described his team as having been "sailing against the wind" and said it had struggled to get the computing power it needed. OpenAI, he argued, had to become a "safety-first AGI company", and he called the building of machines smarter than humans an inherently dangerous endeavour.^[1]

The team is dissolved

OpenAI confirmed on Friday, May 17, 2024, that the Superalignment team had been disbanded, with its members integrated across the company's research efforts.^[1][2][3] OpenAI co-founder John Schulman became the scientific lead for OpenAI's alignment work going forward.^[2]

Reaction

CEO Sam Altman replied to Leike's posts, writing that he was "super appreciative" of Leike's contributions and that Leike was "right we have a lot more to do".^[1]

The two leaders were not the only departures: superalignment members Leopold Aschenbrenner and Pavel Izmailov had also left OpenAI in recent months.^[2]

Why it mattered

The sequence was compressed into a single week: a chief scientist's departure, a public resignation post about safety culture, and the dissolution of the group that was supposed to work on the hardest version of the problem. For people outside the company, it made an internal argument visible.

Every organisation building frontier models carries capability research and safety research in the same body, and the two draw on the same limited pool of researchers, attention and computing power. When a senior researcher says his team was short of compute while products kept shipping, that is a statement about priorities, not only about one team.^[1] Leike's own framing was that building machines smarter than humans is inherently dangerous and needs a company organised around safety first.^[1] Leike did not dispute that there was more to do; so did the chief executive, who said as much in public.^[1]

What the announcement left open is how OpenAI's alignment work is structured from now on: Schulman as scientific lead, and former members of a single named team spread across the company's research efforts rather than gathered in one place.^[2]