OpenAI Reportedly Disbands Its Preparedness Team
OpenAI has reportedly shut down its Preparedness team, the internal unit established in 2023 to evaluate and track risks from frontier AI models that could lead to catastrophic outcomes. The disbandment marks a significant structural change at the company during a period when its models are becoming more capable and more widely deployed.
The Preparedness team was created under a framework that OpenAI published publicly, outlining how the company intended to assess risks across categories including cyberweapons, biological threats, and models acquiring unanticipated autonomous capabilities. Its dissolution raises direct questions about how those evaluation functions will continue — and under whose authority.
This development arrives as OpenAI is scaling its commercial operations aggressively, expanding its enterprise customer base, and releasing increasingly capable reasoning and agentic models. The timing is therefore notable not just internally but for the broader AI safety conversation that OpenAI has helped shape in public policy circles.
The specific functions the Preparedness team held — red-teaming frontier models, maintaining a risk scorecard, and advising on deployment thresholds — were positioned as a structured counterweight to release pressure. It is not yet publicly confirmed what replaces those functions, whether they are redistributed to other internal teams, absorbed into a broader safety organization, or reduced in scope.
For companies and institutions relying on OpenAI's safety commitments as part of their own AI governance posture, the structural question is material. Procurement decisions, regulatory filings, and enterprise AI policies increasingly reference vendor-level safety practices. If OpenAI's internal safety infrastructure is being consolidated or reduced, that changes the baseline assumptions organizations have been working from.
At an industry level, the move will be read through multiple interpretive frames. One is operational efficiency: as AI safety research matures, organizations may be consolidating redundant teams rather than abandoning safety commitments. Another is the more critical read — that commercial pressures are subordinating safety infrastructure at precisely the moment when agentic and autonomous AI systems require more rigorous evaluation, not less.
OpenAI's Preparedness framework was also notable for being one of the few such documents that established specific risk thresholds tied to deployment decisions. Whether those thresholds remain operative under a reorganized structure, or whether they become effectively unenforceable without a dedicated team, is an open and consequential question.
The AIRA assessment is that organizational structure reflects institutional priority. Safety teams are not symbolic — they carry veto-adjacent functions over release timelines, inform board-level risk disclosures, and serve as the internal mechanism through which a lab can credibly claim its deployment decisions are evaluated against harm. When that structure changes, the credibility of the commitments it was designed to enforce changes with it. For enterprises, governments, and partners who have built AI governance assumptions around OpenAI's published safety posture, this is a signal worth tracking carefully — regardless of how the reorganization is eventually characterized.
Sources: — The Verge (https://www.theverge.com/ai-artificial-intelligence/980817/openai-disbands-preparedness-team)