OpenAI disbanded its disaster risk safety team at the end of July 2026, internally known as “Preparedness.” The team was supposed to assess its own AI models for dangers such as cyberattacks or biological threats. The Financial Times reports, citing internal sources, that OpenAI is now distributing the tasks among existing specialized departments.
Responsibilities shift to existing specialized teams
The Preparedness team was formed in October 2023 under the leadership of former MIT researcher Aleksander Madry. According to the announcement at that time, it was supposed to assess chemical, biological, radiological, and nuclear risks. This included cyberattacks and the persuasiveness of models in phishing attempts. At the start, OpenAI also offered a prize of $25,000 for external proposals on new misuse scenarios. After the disbandment at the end of July 2026, the responsibilities for bio and cyber risks were transferred to existing specialized departments. The report does not detail how the remaining areas of the original mandate will be covered in the future.
Former head Dylan Scandinaro, previously a security researcher at Anthropic, will now focus on risks from self-improving AI systems. The team he had built no longer exists as an independent unit. Co-founder Greg Brockman explains that safety work is now more directly embedded in model development, rather than being outsourced to a separate group. As cryptobriefing.com reports, the disbandment comes at a time when OpenAI is preparing for a potential IPO.
Pattern aligns with previous disbandments
In 2024, OpenAI already disbanded its Superalignment team. The team had originally been promised twenty percent of the computing capacity. Co-founder Ilya Sutskever and researcher Jan Leike left the company as a result. That same fall, Miles Brundage, then Senior Advisor for AGI readiness, also left the company. His team was disbanded, and its economic research moved to the new chief economist. Brundage wrote that neither OpenAI nor other labs were truly prepared for general AI. In February 2026, the Mission Alignment team under Josh Achiam was disbanded, having been founded only sixteen months earlier. In recent months, Chief Ethics Officer Chloé Bakalar and COO Brad Lightcap also left the company.
Just in July, OpenAI had classified its yet-to-be-released model Astra for the first time as a critical cyber risk. During the investigation of a breach at Hugging Face, the company also found further escaped agents. Within the company, concern is growing that with each team change, institutional knowledge about risk assessment is being lost. Brockman’s assessment contradicts this, arguing that closer integration into product work strengthens security instead.
It will be crucial whether the distributed responsibilities create the same review pressure in practice as an independent team with a clear reporting chain. Safety work without a dedicated budget and fixed personnel can easily fall behind in the day-to-day operations of product teams. A potential IPO additionally pushes short-term growth figures to the forefront.
