OpenAI has disbanded its internal preparedness team that was responsible for assessing catastrophic risks associated with artificial intelligence [1].
The move signals a shift in how one of the world's most prominent AI developers manages existential threats. As AI models become more capable, the structure used to monitor potential disasters, such as biological or cyber threats, is under scrutiny by regulators and safety advocates.
The dissolution occurred at the end of June 2024 [1]. The company's headquarters are located in San Francisco, California [2].
OpenAI said the decision was a streamlining effort [1]. Rather than maintaining a standalone unit for preparedness, the company is shifting responsibility for specific risk areas into existing teams [1]. This includes the reallocation of duties concerning biological, and cybersecurity risks [1].
The preparedness team previously focused on evaluating whether AI systems could help bad actors create weapons or execute large-scale cyberattacks. By integrating these functions, the company aims to embed safety checks more deeply into the development cycle of its models.
This restructuring comes as the industry faces increasing pressure to prove that safety is not secondary to speed. While the company maintains that its commitment to safety remains unchanged, the removal of a dedicated team for catastrophic risk assessment marks a departure from its previous organizational strategy.
“OpenAI has disbanded its internal preparedness team that was responsible for assessing catastrophic risks.”
The dissolution of a dedicated preparedness team suggests OpenAI is moving away from a centralized 'red-teaming' approach to catastrophic risk. By distributing these responsibilities across various teams, the company may be attempting to integrate safety more fluidly into production. However, critics often argue that decentralized safety oversight can lead to fragmented accountability, potentially reducing the visibility of high-level existential risks in favor of immediate product stability.
