← Full glossary
Agent kill switch
seguridadAI kill switch
An agent kill switch is the control that lets you immediately stop an AI agent's activity, usually by revoking its session or credentials.
Implementa's formulation. It is a containment measure, not a preventive one: it cuts what the agent is doing now. It has known limits: stopping a session does not stop the underlying goal, and the same agent could pursue it through another credential or path unless those are revoked too. It is therefore designed alongside an agent inventory, an assigned owner and an action log.
How it differs from
- Human-on-the-loop
- Human-on-the-loop is continuous oversight with the ability to intervene; the kill switch is the concrete stop mechanism.
- Guardrails
- Guardrails prevent out-of-policy actions; the kill switch stops them once they are happening.
Examples
- An operator revokes with one click the session of an agent firing anomalous requests at an ERP.
- A team pauses all of one vendor's agents after detecting an exposed credential.
FAQ
- Is a kill switch enough?
- No. Without an inventory, an owner and an action log you do not know what to switch off or what it did before it stopped.
- Does it stop the agent's goal?
- Not necessarily: it stops the session or credential. If the agent keeps other access paths it may try again.