one year on
OpenAI disbands Superalignment team after leaders exit, safety researcher says products took priority
The departure of Ilya Sutskever and Jan Leike, combined with the absorption of the team into other groups, raises questions about the company’s commitment to its stated safety goals.
OpenAI has disbanded its Superalignment team, the group dedicated to steering and controlling artificial intelligence systems smarter than humans, less than a year after its formation. The dissolution follows the departures of both co-leads: chief scientist and co-founder Ilya Sutskever, who left the company Tuesday, and Jan Leike, a former DeepMind researcher who resigned the same day. A person familiar with the situation said some team members are being reassigned to other research groups.
Leike elaborated on his exit in a thread on X on Friday, writing, “I have been disagreeing with OpenAI leadership about the company’s core priorities for quite some time, until we finally reached a breaking point.” He said his team struggled for computing resources and that “safety culture and processes have taken a backseat to shiny products.” The Superalignment team had been promised 20 percent of OpenAI’s computing power over four years when it was announced last July. Leike said his team struggled for computing resources.
The news comes days after OpenAI unveiled GPT-4o, a faster multimodal model. The team departures have amplified external criticism of OpenAI’s priorities and reignited debates about whether the company can balance rapid product release with its founding mission of safe AGI.
The record
Posted on X that he disagreed with OpenAI leadership about core priorities for some time and that his team struggled for computing resources; said “safety culture and processes have taken a backseat to shiny products.”
Responded on X saying he was sad to see Leike leave and that the company had more work to do, while also praising Ilya Sutskever as “one of the greatest minds of our generation.”
Posted a joint statement with Altman on X asserting that OpenAI has raised awareness of AGI risks and opportunities so the world can better prepare.
One year later — open only if you can handle spoilers
In the months that followed, several more safety researchers left OpenAI, and the company continued to face scrutiny over its commitment to safety. The Superalignment team was not reconstituted, and OpenAI has not publicly recommitted to allocating 20% of compute to safety research.