OpenAI disbanded the Preparedness team for assessing catastrophic AI risks
OpenAI abolished the Preparedness team at the end of July 2026. The team assessed biological and cyber risks posed by its AI models. Existing departments took over the work, while several safety employees also left the company.
OpenAI disbanded the Preparedness team at the end of July 2026. The team assessed whether AI models from the company could pose serious or catastrophic risks. According to Financial Times, which cites internal sources, the work of the team focused on biological and cyber risks was divided among existing departments.
Former team leader Dylan Scandinaro now focuses on safety risks associated with so-called recursively self-improving AI, meaning systems that can optimize themselves and train other models. Company co-founder Greg Brockman said that OpenAI had linked safety work more closely to model development itself.
Several employees working on safety have recently left the company, including chief ethics officer Chloe Bakalar and employee Joshua Achiam. According to one internal source, employees increasingly feel a sense of responsibility and concern that OpenAI is not doing enough for safety. Employees have also spoken publicly about the issue, particularly after an autonomous hacking incident involving Hugging Face — one employee expressed hope that the company would treat it as a warning sign.
Why it matters
The abolition of a dedicated team for catastrophic risks at one of the leading AI labs shows how the organization of safety oversight is changing in practice — the assessment of biological and cyber risks no longer has a separate independent team but is part of routine development. According to internal sources, this is accompanied by departures of safety employees and growing concerns within the company, making it a relevant signal for anyone assessing the safety practices of AI model suppliers.
Relevant practical impact
What this means
For a business
Companies relying on models from OpenAI are losing a dedicated external point of oversight that assessed biological and cyber risks posed by models at the supplier — this work is now part of regular development teams, changing the nature (though not necessarily the scope) of safety oversight at a key AI supplier.
Risks and complianceCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.