OpenAI disbanded its Preparedness team at the end of July 2026, according to the Financial Times, and reassigned biological and cyber risk assessment to existing teams, eliminating a centralized unit whose job was to determine whether models could be misused for severe harms such as bioweapons development or large-scale cyberattacks before release. The move follows the company's earlier decisions to dissolve its AGI readiness and superalignment teams, reinforcing concerns among critics that OpenAI has been shrinking the standalone safety infrastructure it once emphasized even as its models become more powerful. Dylan Scandinaro, who led the Preparedness team after joining from Anthropic in February, remains at OpenAI and is now focused on the implications of recursive self-improving AI. The restructuring has also coincided with recent departures from OpenAI's safety and ethics ranks, including Chloé Bakalar, Josh Achiam and Johannes Heidecke. Jan Leike, who left OpenAI in 2024, told the Financial Times that the company was prioritizing "shiny products" over safety. The changes come as OpenAI is expected to move toward a major IPO, a backdrop that has intensified scrutiny over how the company will measure and maintain accountability after folding dedicated risk oversight into broader teams.