Reports citing the Financial Times say the OpenAI preparedness team disbanded at the end of July, with the company moving responsibility for specific AI-risk areas into existing groups. The change matters because the unit assessed whether OpenAI’s models could create severe or catastrophic risks and developed ways to reduce them.
The account remains unconfirmed by OpenAI in the material available. The Verge, India Today and Engadget each reported the development through the Financial Times, rather than a direct company announcement.
What did OpenAI’s preparedness team do?
The team’s job was to evaluate whether OpenAI models posed serious risks and devise mitigations, according to The Verge and India Today. Those risks included work categorized around bio and cyber capabilities.
Under the reported reorganization, senior staff or responsibilities for those areas were assigned to existing teams. That distinction is doing real work: the reporting describes the end of a centralized preparedness unit, not the elimination of all work on model safety.
India Today reported that layoffs were not thought likely to be part of the change. Engadget described the move as part of a restructuring and reported that OpenAI called the cuts a streamlining effort ahead of an initial public offering, following an internal push to focus on ChatGPT.
Who led the team, and what happens next?
Dylan Scandinaro, the preparedness team’s head, joined OpenAI from Anthropic in February, The Verge reported. He has reportedly shifted to examining the safety implications of “recursive self-improving” AI.
That term refers to AI systems improving themselves. The supplied reporting does not spell out Scandinaro’s new remit beyond its focus on the safety implications of that capability.
The reported closure follows other changes to OpenAI’s safety organization. India Today said the company had previously dismantled groups focused on AGI readiness, mission alignment and superalignment. Engadget also reported recent departures by ethics lead Chloé Bakalar and safety head Johannes Heidecke.
Those moves have drawn criticism from some observers. Jan Leike, a former OpenAI employee who resigned in 2024, told the Financial Times that the company was prioritizing “shiny products,” The Verge reported. That is Leike’s assessment, not a conclusion established by the reorganization itself.
What the reports establish is narrower: OpenAI appears to have replaced one dedicated preparedness team with responsibility distributed across existing teams. Whether that structure provides more, less or equivalent scrutiny will depend on how those teams are staffed and run, details not included in the reports.
This story draws on original reporting from The Verge.
