OpenAI introduces Admin plugin for ChatGPT Work and Codex

OpenAI has introduced an Admin plugin for ChatGPT Work and Codex, giving workspace administrators a conversational way to inspect activity, change settings and carry out supported management tasks. OpenAI announced the plugin on August 25, 2026. The company describes it as a connection between questions, workspace context and authorised actions: an admin can ask about usage, investigate details, make a permitted change and receive a structured result in the same conversation. The plugin covers several routine areas. Administrators can review activity and credit use across ChatGPT Work and Codex, identify members or groups nearing limits, and see where additional enablement may be useful. They can add or remove members, update groups, handle onboarding and offboarding, review effective permissions, and control feature or model access by role or group. The plugin also supports usage-limit and spending requests, including approving or denying requests with current usage as context. OpenAI says recurring checks can be automated as well. Pending usage requests can be routed to Slack or Microsoft Teams for authorised review. Feature-access requests can be monitored and granted automatically when they satisfy predefined criteria, with exceptions sent to a human. The announcement stresses that the plugin does not expand an admin’s existing authority. Each request is mapped to a supported read or write action within the user’s role, workspace policies and approval requirements. Admins can review what they asked for, whether it completed and what changed; broader-impact changes can be reviewed before application. The news matters because administration is becoming part of the AI work surface itself. For companies, that could reduce the switching between reports, settings and separate consoles involved in routine operations. It also makes governance more important: organisations still need clear roles, spending rules, audit processes and review points for consequential changes. OpenAI cites its own IT team as an example. It says ChatGPT Work agents in Slack resolved about 45 percent of ticket volume at the time of reporting, while support demand roughly doubled. Those figures are OpenAI’s account, not an independent evaluation. The practical takeaway is a permission-aware admin interface, not unrestricted automation. Teams will need to verify availability, supported actions and their own approval controls before relying on it in production.