Jan Leike resigns and the superalignment team dissolves

Jan Leike, co-lead of OpenAI’s Superalignment team, announced his resignation in a thread on X, three days after chief scientist Ilya Sutskever announced he was leaving the company he had co-founded. Leike’s last working day had been 16 May; he posted the following morning.

Leike’s thread was pointed. He wrote that he had “been disagreeing with OpenAI leadership about the company’s core priorities for quite some time, until we finally reached a breaking point,” and that “over the past years, safety culture and processes have taken a backseat to shiny products.” He described his team as having been “sailing against the wind,” struggling for the computing resources it needed even though OpenAI had originally committed 20% of its compute to superalignment work when the team launched in July 2023. He argued that building smarter-than-human systems was “an inherently dangerous endeavor” and that OpenAI needed to become “a safety-first AGI company” in substance, not just in messaging.

OpenAI did not appoint new co-leads. The Superalignment team, formed to solve the technical problem of controlling AI systems more capable than their supervisors within four years, was effectively dissolved, with remaining researchers redistributed across other teams. Sam Altman responded on X that he was “super appreciative” of Leike’s contributions and “very sad to see him leave,” promising a longer statement; OpenAI’s public follow-through was the board’s creation, eleven days later, of a Safety and Security Committee whose members included Altman himself, which drew criticism for seating the chief executive on the body meant to check him.

Leike joined Anthropic within two weeks to continue alignment research there, and the episode became a reference point in subsequent arguments over whether frontier labs’ internal safety commitments could survive contact with product and revenue pressure.