OpenAI's entire Superalignment team, which was focused on the existential dangers of AI, has either resigned or been absorbed into other research groups
The entire OpenAI team focused on the existential dangers of AI has either resigned or been absorbed into other research groups, WIRED has confirmed.
Context & Ripple Effects
OpenAI created Superalignment as a dedicated effort to steer and control more capable systems, with compute explicitly earmarked for that work in its initial team launch. Days before this report, co-lead Jan Leike had resigned, removing one of the unit's central leaders amid the leadership departures.
The team’s disappearance as a standalone organization matters because it changes how visibly and independently OpenAI’s long-horizon safety work is organized, even where researchers continue in other groups.
First-order effects
- OpenAI no longer has a distinct Superalignment team; its remaining personnel are either gone or distributed across other research organizations.
- The departures and reorganization reduce the unit’s dedicated leadership and make its original, separate mandate harder to assess externally.
Second-order effects
- OpenAI’s safety claims face greater scrutiny because work once concentrated in a named group is now embedded in broader research operations.
- Other frontier-AI labs have an incentive to show whether their own safety functions have independent authority, dedicated resources, and retention of senior researchers.
Third-order effects
- If dedicated alignment teams are repeatedly folded into core research groups, frontier-AI governance may shift from visibly separate safety institutions toward internal operational processes that are less legible to outside stakeholders.
- That shift could make organizational design—reporting lines, resource commitments, and researcher autonomy—a more important measure of safety capacity than a lab’s public team names.
The trend: This is a data point in the institutionalization of frontier-AI safety, where alignment work increasingly competes with and becomes embedded in the labs building the models.