Report: OpenAI Dissolves Preparedness Team, Dispersing Safety Functions Across Existing Departments

According to The Financial Times, OpenAI disbanded its Preparedness team in late July and transferred its previous responsibilities to existing company teams across specific areas such as biosecurity and cybersecurity. This team had previously been responsible for assessing whether frontier models could cause severe harm and researching corresponding risk mitigation measures. This adjustment is part of ongoing restructuring within OpenAI’s safety research organization over recent years. The company has already dissolved teams including Superalignment and AGI Readiness. Recently, Chloé Bakalar, Head of AI Ethics, Josh Achiam, Chief Futurist, and Johannes Heidecke, Head of Safety Systems, have also departed successively. Jan Leike, former head of Superalignment who left OpenAI in 2024, criticized that the company was neglecting safety work in favor of “flashy new products.” However, this statement reflects the perspective of departing employees and should not be taken as definitive evidence that OpenAI has abandoned its investment in safety. Dylan Scandinaro, head of the Preparedness team, joined OpenAI from Anthropic earlier this year. Following the dissolution of his team, he will shift his focus to studying the potential impacts of recursively self-improving AI—that is, the capabilities and risks arising from AI systems participating in the development and improvement of subsequent AI systems. These organizational changes coincide with OpenAI’s preparations for a potential IPO, but there is currently no evidence indicating a direct causal relationship between the two events.