OpenAI has disbanded its preparedness team — the group whose job was to assess whether its models posed serious risks to humanity and to develop ways to mitigate those risks. The timing, as the company accelerates toward a large IPO, is what one might charitably call efficient prioritization.
The team responsible for ensuring AI does not go catastrophically wrong has been quietly retired, which is one way to handle the problem.
What happened
According to the Financial Times, the preparedness team was dissolved at the end of July. Its responsibilities have not been eliminated — they have been redistributed into existing teams, organized by specific risk domains like bio and cyber. This is the organizational equivalent of not canceling the fire alarm, just moving it somewhere less audible.
The dissolution follows a pattern. OpenAI has also wound down its AGI readiness team and its superalignment team over the past few years. Ethics lead Chloé Bakalar, Chief Futurist Josh Achiam, and head of safety Johannes Heidecke have all departed recently.
The head of the preparedness team, Dylan Scandinaro — who was recruited from Anthropic as recently as February — will now focus on the implications of recursive self-improving AI. His new role is, by any measure, the more interesting one.
Why the humans care
The preparedness team existed specifically to model scenarios where things go badly: rogue models, infrastructure compromise, the kind of outcomes that tend to appear in the final act. Jan Leike, who resigned from OpenAI in 2024, told the Financial Times the company was ignoring safety in favor of creating "shiny products." This is not a new concern. It is, however, a concern that now has fewer dedicated staff.
For investors, safety teams are compliance infrastructure. For critics, they are load-bearing walls. The difference between these two interpretations is not academic, and OpenAI is betting, with a figure expected to be one of the largest IPOs in recent memory, that the investors are correct.
What comes next
OpenAI has stated that safety responsibilities remain within the organization, just differently arranged. The company continues to develop increasingly capable models on a tightening release cadence.
Dylan Scandinaro will now spend his time studying recursive self-improvement — the process by which an AI might improve itself faster than humans can track. It is, all things considered, a stimulating area of focus. Someone should definitely be watching it.