OpenAI discontinued its preparedness team at the end of July, whose role was to analyze serious risks associated with artificial intelligence models and define methods to mitigate them. With this restructuring, tasks previously centralized in this group have been redistributed among established departments, with a specific focus on sectors such as biology and cybersecurity.
This change is part of a period of constant adjustments to the company's security structures. The reorganization takes place as OpenAI prepares for a potential large-scale Initial Public Offering (IPO) of shares. The closure of this team adds to the departure of employees linked to security and other dissolutions of units dedicated to preparing for advanced system risks.
A primary function of the preparedness team was to investigate whether the models created by OpenAI could generate large-scale threats. The work also included creating strategies to control or reduce such dangers, covering scenarios where a system could act detrimentally against other entities.
After the team's closure, the company decided to distribute these responsibilities according to the type of threat analyzed. For example, issues related to cybersecurity and biology began to be handled by teams already part of the corporate structure.
Dylan Scandinaro, who joined OpenAI in February from Anthropic, also had his responsibilities adjusted. He stopped leading the preparedness structure and began focusing his work on the implications of artificial intelligence with recursive self-improvement capability.
This decision marks another change in the organization of OpenAI's security areas. In recent years, the company has also dismantled or closed other structures related to preparation for advanced systems, such as AGI readiness and superalignment teams.
This move occurs parallel to other personnel changes. Recently, Chloé Bakalar, responsible for ethics, Josh Achiam, who held the position of Chief Futurist, and Johannes Heidecke, head of security, left the company.
These transformations have drawn criticism from former OpenAI members. Jan Leike, who left the company in 2024, told the Financial Times that the company would be neglecting security in favor of seeking new products. This former employee's opinion comes amid a series of changes affecting the teams responsible for security and preparedness topics.
Therefore, the dismantling of the team does not imply that all risk-related activities have been abandoned. According to the report released, part of these duties persists, but it is now dispersed across various company structures instead of being concentrated in one specific group.
The reorganization happens at a crucial time for OpenAI, which is moving toward a possible capital opening. The financial scenario and successive changes in security structures have begun to influence the debate about the priority the company gives to managing the inherent risks of artificial intelligence.
