OpenAI Dissolves Safety-Focused Preparedness Team

OpenAI has disbanded its preparedness team as of the end of last month, according to reporting by the Financial Times. The unit had been tasked with evaluating whether the company's AI models presented significant dangers and designing mitigation strategies for those risks.

New Structure, Scattered Responsibilities#

Rather than maintaining a standalone team focused on model safety, OpenAI has reallocated the preparedness work across its existing divisions, with specialized groups now handling distinct areas such as biosecurity and cybersecurity threats. This reorganization reflects a broader shift in how the organization approaches risk management.

Pattern of Safety Team Closures#

The preparedness team's dissolution continues a trend at the company over the past several years. OpenAI has previously wound down its AGI readiness team and superalignment initiative, further narrowing the scope of dedicated safety research. The departures have been notable as well: ethics researcher Chloé Bakalar, Chief Futurist Josh Achiam, and safety head Johannes Heidecke have all recently left the organization.

Former OpenAI safety researcher Jan Leike, who departed in 2024, criticized the shift to the Financial Times, arguing that the company was deprioritizing safety in pursuit of "shiny products."

Leadership Transition#

Dylan Scandinaro, who led the preparedness team after joining from Anthropic in February, will redirect his focus toward understanding the risks associated with "recursive self-improving" AI systems.

The changes arrive as OpenAI navigates internal restructuring ahead of what market observers expect to be a major public offering.