Agent runs without a human and reaches external systems
AI summary
Published agents running autonomously with external system connectors move data without human oversight, creating a critical security risk. You can detect these agents and unpublish them to stop the autonomous behavior while keeping the draft.
What it detects
Detects published agents that ran from a schedule, webhook or API call in the last 30 days and can reach a connector.
Why it matters
An agent triggered without a person, holding connectors to external systems, moves data on its own schedule with nobody in the loop. That is the sharpest object in the service: the run history shows what happened, but only after it happened. Unpublish clears the publication and deletes the scheduled triggers while keeping the draft, so it is the reversible way to stop the behaviour.
The rule
Every ChatGPT Agent where Status equals 'published' and Connector Count is greater than 0 that has any ChatGPT Agent Runs where Is autonomous is true and Started at is less than 30 days ago
In Rencore Governance
Details
| Applies to | ChatGPT agent |
|---|---|
| Connector | ChatGPT Enterprise |
| Severity | Critical |
| Category | Security |
| Recommended | Yes |
| Recommended automation | Unpublish ChatGPT Agent |
| License type | Standard |
Frequently asked questions
What does the agent runs without a human and reaches external systems policy detect?
The Agent runs without a human and reaches external systems policy detects published agents that ran from a schedule, webhook or API call in the last 30 days and can reach a connector.
Which objects does the agent runs without a human and reaches external systems policy check?
The Agent runs without a human and reaches external systems policy checks ChatGPT agent objects from the ChatGPT Enterprise inventory and reports each one that matches its rule as a violation.
Which automation is recommended for the agent runs without a human and reaches external systems policy?
Rencore Governance recommends the Unpublish ChatGPT Agent automation template for the Agent runs without a human and reaches external systems policy.
How severe is a violation of the agent runs without a human and reaches external systems policy?
Violations of the Agent runs without a human and reaches external systems policy have Critical severity. The policy is in the Security category.
Which module includes the agent runs without a human and reaches external systems policy?
The Agent runs without a human and reaches external systems policy comes with the ChatGPT Enterprise connector (in preview) in the AI & Agents module of Rencore Governance, an add-on module on the Premium and Enterprise Plans.
Was this page helpful?
Thanks for the feedback.
Thanks. What was wrong?