Commissioned by Kiteworks
Stop an AI agent

Inside the CTRL Vault · not a film still

Stop an AI agent

How can organizations stop an AI agent?

Cutting off data access and shutting down the agent are separate controls. Test both.

Use two separate controls: revoke the agent’s data access and shut down its running software. Cutting off one access route does not necessarily stop activity through other tools or undo earlier transfers. Decide who can activate each control, test both before an incident and keep evidence that access was denied and execution ended.

Confirm that each relevant access path uses the Secure MCP Server’s enforcement point, then inspect a denied request after revocation. Its stated scope is Kiteworks-governed content; Kiteworks says the server does not inspect agent coordination elsewhere. A denied request therefore does not prove that other tools stopped or that earlier transfers were undone. Check those other tools and access paths separately.

Document who can revoke access and who can invoke runtime shutdown, including how you will observe termination. Use a non-production agent to test both controls before an incident and measure their results. Preserve evidence that governed access was denied and execution ended; a revoked-credential entry alone proves neither.

Put this into practice

  • Assign authority for access revocation and runtime shutdown.
  • Test the next governed request after revocation.
  • Test runtime termination separately and retain evidence of both outcomes.

Sources

Bring it to a demo

See the controls in your workflow.

Bring one AI task and the data it needs. Ask Kiteworks to show what is allowed, what is denied and the record of each decision.