OpenAI has ended its dedicated "Superalignment" team, according to Bloomberg reporting cited by The Decoder, and will fold that work into broader safety research. The move matters because the team was focused on the safety of future high-performance AI systems, a topic now sitting at the center of debate inside and outside the company.
What changed inside OpenAI
The clearest fact is structural: the Superalignment team no longer exists as a separate unit. Its mission, as described in the source, was to help ensure the safety of future high-performance AI systems. OpenAI is not abandoning safety work altogether; instead, that work is being integrated into general safety research.
That distinction is important. A dedicated team can give one problem a clear identity, a budget, and a mandate. Folding the work into a wider research operation may make safety more distributed across the company, but it can also make it harder for outsiders to see who is accountable for the most difficult questions.
The timing has intensified the scrutiny. Former safety researcher Jan Leike left OpenAI and sharply criticized the company's leadership for not taking the risks of very advanced AI systems seriously. After that departure, OpenAI CEO Sam Altman and co-founder and President Greg Brockman published a lengthy statement on X that The Decoder characterized as ultimately vague.
The AGI risk debate is not settled
One possible explanation raised in the source is that OpenAI may have concluded that AGI risk had been overestimated. Under that theory, safety research focused specifically on super-AI may have grown beyond what management thought was justified, leading to a reduction of resources and the end of the separate team.
That reading would fit the basic facts of the Superalignment team's elimination. It would also help explain why some safety-focused staff might feel misunderstood and decide to leave. But the source also points to evidence that complicates this interpretation.
John Schulman, co-creator of ChatGPT and co-founder of OpenAI, is expected to take over Leike's role. His recent comments do not suggest that AGI should be dismissed as a distant or irrelevant concern. Schulman believes that AGI will be possible within the next two to three years.
He has also suggested cross-organizational rules, including a possible pause, to avoid rolling out such a system to many people without clear safety rules. That position keeps the safety question alive even as OpenAI changes the way it organizes the people working on it.
"If AGI came way sooner than expected, we would definitely want to be careful about it. We might want to slow down a little bit on training and deployment until we're pretty sure we know we can deal with it safely," Schulman says.
Critics see different problems
The source highlights a sharply different view from Meta's chief AI researcher Yann LeCun, who is dismissive of near-term AGI. His criticism is not aimed only at technical forecasts, but also at the culture that can form around fears of a powerful future system.
"When you put sufficiently many people in a room together with such a distorted view of reality that they perceive an impending Great Evil, they often fall victim to a spiral of purity that makes them hold more and more extreme beliefs. Pretty soon, they become toxic to the organization that hosts and funds them. They become marginalized and eventually leave," writes Meta's chief AI researcher Yann LeCun.
This contrast shows why the Superalignment decision is difficult to interpret from the outside. To some observers, it may look like a company correcting an overreaction to speculative risk. To others, it may look like the removal of a focused safety function just as more advanced systems become plausible.
The available facts do not resolve that dispute. They do show that OpenAI is trying to keep AI safety in its official message while changing the team structure that had been associated with its most ambitious safety concerns.
Trust now depends on governance as much as research
The article also points to a separate issue that affects trust in OpenAI's leadership: employee departure agreements. According to Altman on X, he said he did not know about gag clauses that OpenAI employees had to sign when they left the company.
One clause threatened employees with the potential loss of millions of dollars in OpenAI stock if they spoke critically about the company after leaving. The source says that clause was uncovered by a journalist at Vox.
"Although we never clawed anything back, it should never have been something we had in any documents or communication. This is on me and one of the few times I've been genuinely embarrassed running OpenAI; I did not know this was happening, and I should have," Altman writes.
That admission matters because AI safety is not only a technical research problem. It also depends on whether researchers can raise concerns, whether leadership hears them, and whether the company has rules that encourage or discourage public criticism after employees leave.
The Superalignment team's closure, Leike's criticism, Schulman's cautious comments, LeCun's dismissal of near-term AGI fears, and Altman's statement about gag clauses all point to the same larger issue: OpenAI is under pressure to explain how it will manage the safety of very advanced AI systems while also maintaining confidence in its internal decision-making.
What to watch next
The immediate question is how general safety research will absorb the former Superalignment work. If the same concerns remain active, OpenAI will need to show how they are being handled without a team carrying that specific name.
Several issues now define the debate:
- whether AGI is close enough to require special rules now;
- whether safety work is stronger as a dedicated unit or as part of general research;
- whether employees can challenge leadership decisions without fear of professional or financial consequences;
- whether OpenAI can explain its safety approach with enough clarity to satisfy critics.
For readers following AI safety, the key point is not simply that one team was disbanded. It is that OpenAI is reorganizing a sensitive part of its safety work while former staff, company leaders, and outside researchers disagree sharply about the scale and urgency of AGI risk.