OpenAI dissolved its preparedness team at the end of last month, according to the Financial Times. The unit's mandate was to evaluate whether AI models posed serious risks and develop mitigation strategies against potential dangers, including scenarios where models might act autonomously to compromise external systems.
The disbanded team had operated as a dedicated safety function tasked with stress-testing OpenAI's models before deployment. This included assessing risks from model capabilities that could enable harmful behavior if misused or if the systems operated outside their intended parameters.
Responsibility for preparedness work now falls to other OpenAI divisions. The FT report did not specify which teams absorbed the function or how the reorganization affects the company's overall approach to risk assessment. This shift comes as major AI labs face increasing scrutiny over safety practices, particularly around large language models and their potential for misuse.
The decision reflects a broader pattern within OpenAI of prioritizing product development and scaling. The company has faced criticism from safety researchers over the past year for what some view as an acceleration in capabilities work without proportional investment in safety mechanisms. The preparedness team's dissolution removes a dedicated team focused specifically on worst-case scenarios and emergent risks.
OpenAI has not publicly commented on the reorganization. The company previously emphasized the importance of preparedness work, with executives acknowledging that identifying risks before they manifest in production systems forms a core part of responsible deployment. The shift to distributing preparedness responsibilities across multiple teams could mean less coordinated oversight of systemic risks or simply a structural reorganization of existing safety work.