OpenAI Disbands Preparedness Team, Report Says

Preparedness work concerns evaluating whether increasingly capable models could enable severe harms and identifying measures to reduce those risks before deployment. The Verge's description specifically frames the former team's remit around both risk assessment and mitigation development.

The reported disbandment does not, on its own, establish whether the underlying evaluations have ended, been consolidated elsewhere, or changed in scope. The available reporting states that responsibility was reassigned, but the supplied article text does not identify the receiving group.

For ML practitioners, the operational question is not simply whether a named safety team exists, but how evaluation ownership, escalation authority, and deployment gates are implemented. Companies conducting comparable frontier-model work often need clear accountability for threat modeling, capability evaluations, red-team findings, and mitigation verification across research and product organizations.

A reorganization can preserve those controls if responsibilities, decision rights, and audit trails remain explicit. Conversely, industry discussion around similar changes frequently focuses on whether independent review and consistent evaluation criteria survive when functions are distributed across teams. The current reports provide no evidence sufficient to determine those details for OpenAI.