OpenAI’s work on controlling future “superintelligent” AI systems is facing renewed scrutiny after key departures, reported resource constraints and a shift away from a dedicated Superalignment team.
The issue is not simply organizational. At stake is how much priority the company gives to safety research while it continues building and launching increasingly capable AI systems.
The reported gap between promise and practice
OpenAI created the Superalignment team last July with a large stated mission: to solve the core technical challenges of controlling superintelligent AI in the next four years. The team was responsible for research aimed at governing and steering systems more powerful than those available today.
According to a person from that team, it had been promised 20% of OpenAI’s compute resources. But the same source said requests for only a fraction of that compute were often denied, blocking the group from doing its work.
That matters because compute is not a side detail in AI research. For a team trying to test, evaluate and understand advanced systems, access to the company’s technical resources can determine what research is possible, how quickly it happens and whether safety work keeps pace with product development.
The reported compute issue was one of several concerns that pushed multiple members of the team to resign. Among them was Jan Leike, the team’s co-lead and a former DeepMind researcher who worked at OpenAI on ChatGPT, GPT-4 and ChatGPT’s predecessor, InstructGPT.
Why the Superalignment team mattered
The Superalignment team was led by Leike and OpenAI co-founder Ilya Sutskever, who also resigned from the company this week. It brought together scientists and engineers from OpenAI’s earlier alignment division, along with researchers from other parts of the company.
Its brief extended beyond internal work. The team was expected to contribute research that could inform the safety of both OpenAI models and non-OpenAI models. It also used initiatives including a research grant program to solicit work from the wider AI industry and share findings outside the company.
The group did produce safety research and distributed millions of dollars in grants to outside researchers. But as product launches took up more of OpenAI leadership’s attention, the team increasingly had to push for upfront investment it considered critical to OpenAI’s mission of developing superintelligent AI for the benefit of all humanity.
Leike publicly described a disagreement with OpenAI leadership over priorities. He said more attention should go toward areas including security, monitoring, preparedness, safety, adversarial robustness, alignment, confidentiality and societal impact. His concern was that the company was not on a trajectory to handle those challenges properly.
The leadership turmoil around the work
The Superalignment team was also affected by broader internal disruption at OpenAI. Sutskever’s conflict with OpenAI CEO Sam Altman became a major distraction for the company.
Late last year, Sutskever and OpenAI’s old board of directors moved to abruptly fire Altman over concerns that Altman had not been “consistently candid” with board members. Under pressure from investors, including Microsoft, and many OpenAI employees, Altman was eventually reinstated. Much of the board resigned, and Sutskever reportedly never returned to work.
According to the source, Sutskever was central to the Superalignment team’s influence inside OpenAI. He contributed research, connected the group to other divisions and helped communicate the importance of its work to key decision-makers.
That role is difficult to replace. A safety team can publish research, but it also needs internal standing if its conclusions are going to shape what a fast-moving AI company builds, delays, tests or changes.
What replaces the dedicated team
After Leike’s departure, Altman wrote on X that he agreed there was “a lot more to do” and that OpenAI was “committed to doing it.” OpenAI co-founder Greg Brockman later added that the company needs rigorous testing, careful consideration, world-class security and a harmony of safety and capabilities.
Even so, the structure is changing. John Schulman, another OpenAI co-founder, has moved to lead the kind of work the Superalignment team had been doing. But there will no longer be a dedicated team with the same shape.
Instead, the work will be handled by a loosely associated group of researchers embedded across OpenAI divisions. An OpenAI spokesperson described the change as “integrating [the team] more deeply.”
That approach could put safety researchers closer to product and model teams. But it also raises the concern that a dedicated safety agenda may lose visibility, independence or leverage when spread across the company.
The unresolved question for OpenAI
The core concern is whether OpenAI’s AI development will be as safety-focused as it could have been with a dedicated Superalignment team, clear leadership and reliable access to promised resources.
The company’s stated goal involved solving hard technical problems before future systems become harder to control. The reported friction shows how difficult that becomes when safety work competes with product launches, leadership attention and internal politics.
For OpenAI, the next challenge is not only to say that safety remains important. It is to show how safety research will receive enough resources, authority and continuity inside a company building systems it describes as increasingly powerful.