OpenAI has disbanded its preparedness team, the group tasked with assessing whether its models posed serious risks and developing methods to mitigate them. The change reportedly took place at the end of last month.
The preparedness team examined scenarios in which a model might behave dangerously, including the possibility that an AI system could go rogue and hack another company. Responsibility for evaluating these threats has now been divided across specific areas such as biological and cyber risks, with the work moved into existing teams within the organisation.
A Company in Transition
The move is the latest in a series of significant changes at OpenAI, which has faced considerable upheaval as it heads towards what is expected to be a substantial initial public offering. Over the past few years, the company has gradually dismantled its more research-led structure, dissolving both its AGI readiness and superalignment teams.
Several senior figures have departed recently. Ethics lead Chloé Bakalar, Chief Futurist Josh Achiam, and head of safety Johannes Heidecke have all left the company. Their exits have drawn criticism, with Jan Leike, who resigned from OpenAI in 2024, saying the company was prioritising the creation of “shiny products” over safety.
New Focus for Former Team Lead
Dylan Scandinaro, who led the preparedness team and was recruited from Anthropic in February, will now concentrate on the implications of “recursive self-improving” AI. This form of technology refers to systems capable of enhancing their own capabilities.
The restructuring reflects a broader shift in how OpenAI approaches long-term AI safety, moving away from dedicated specialist teams and integrating risk assessment into its wider operational structure. The redistribution of responsibilities took effect as the company continues to prepare for its anticipated market debut.
Source
Image: theverge.com
