OpenAI dismantles safety team ahead of IPO

OpenAI has dismantled its preparedness team—the unit tasked with evaluating existential risks from advanced AI models—reported the Financial Times on Sunday. The team, led by Dylan Scandinaro (poached from Anthropic in February), previously focused on risks like rogue AI systems or biosecurity threats. Instead of centralizing safety oversight, OpenAI is redistributing risk assessment duties to specialized groups handling areas like cybersecurity and bioengineering. This restructuring comes as the company accelerates toward a widely anticipated IPO, a milestone that has coincided with a series of high-profile departures from its safety ranks.

Critics like former OpenAI researcher Jan Leike argue the company is deprioritizing safety in favor of rapid product development. Leike, who resigned in 2024 and has since criticized OpenAI’s direction, told the FT that the company’s actions reflect a broader trend of dissolving core safety teams, including its AGI readiness and superalignment initiatives. Key leaders in ethics and futurism—Chloé Bakalar, Josh Achiam, and Johannes Heidecke—have also left recently, raising questions about OpenAI’s commitment to long-term risk mitigation as it races toward public markets.