OpenAI has reportedly shut down the team responsible for assessing whether its models posed serious risks. The move is the latest shakeup of the company’s safety work as it heads toward an IPO.
According to the Financial Times, OpenAI disbanded its preparedness team at the end of last month. The group’s job was to study serious model risks and find ways to reduce them.
That responsibility is now split across areas like biological and cyber threats. The work has been folded into existing teams instead of living under one roof.
The change follows years of retreat from OpenAI’s earlier safety structure. The company has already dissolved its AGI readiness and superalignment teams. Several leaders have left, including ethics lead Chloé Bakalar, chief futurist Josh Achiam, and head of safety Johannes Heidecke.
Critics see a clear pattern. Jan Leike, who resigned from OpenAI in 2024, told the Financial Times the company was chasing “shiny products” instead of safety.
The preparedness team’s leader, Dylan Scandinaro, will stay at the company. Scandinaro, who joined from Anthropic in February, will now focus on the risks of “recursive self-improving” AI.
The shift matters because OpenAI’s models keep getting more capable. Its Astra model recently solved ten long-open math problems, as we covered. Safety teams are meant to watch for what a more powerful model might do next.
As OpenAI races toward a public listing, the question is whether safety work can keep pace with the products it ships.
Source: The Verge (citing the Financial Times). Related: Rogue AI Is No Longer Science Fiction.
