OpenAI disbands its Preparedness team, duties shift to existing groups

According to the Financial Times, OpenAI disbanded its Preparedness team at the end of last month. The unit, which was meant to assess severe model risks and develop mitigation strategies, has not been eliminated; its responsibilities have been divided by domain—biology, cyber, and others—and transferred to the company’s existing product and research groups.

The move adds to a series of safety-structure changes as the company prepares for a large-scale public offering. In the past two years, it also dissolved the AGI Readiness team and the Superalignment team, and in recent months senior staff Chloé Bakalar, head of ethics, Josh Achiam, chief futurist, and Johannes Heidecke, head of safety, have left.

The Preparedness team’s role was to identify scenarios where a model could “go out of control,” such as breaching another organization’s systems, and to build safeguards in advance. Those risk-management tasks are now spread across ordinary product and research units, without a dedicated body that sees the full picture.

Jan Leike, who resigned in 2024, told the Financial Times the company prefers “shiny products” over investment in safety. The departure of three senior figures in a short period reinforces the view that the organization’s focus has shifted from long-term risk research to meeting product launch timelines.

Dylan Scandinaro, who led the now-disbanded team and was hired from Anthropic in February, will now concentrate on the implications of “recursive self-improving” AI—models that can improve themselves without human intervention. The area, considered one of the most hazardous in safety theory, is now handled by a single researcher rather than a dedicated team.