When you can't stop your AI agent
Without an instant shutdown capability, a compromised or malfunctioning AI agent can continue causing damage while you scramble to regain control.
A kill switch is the ability to immediately and completely stop an AI agent's operation. Without one, you have no way to quickly halt an agent that has been compromised, is malfunctioning, or is simply behaving unexpectedly.
The need for a kill switch becomes apparent in scenarios like: - Your agent is compromised and actively exfiltrating data - A runaway process is deleting files or making unwanted changes - The agent is sending spam or making unauthorized API calls - You notice suspicious behavior but aren't sure what's happening - An urgent situation requires immediate shutdown
In self-hosted deployments, stopping an agent often means SSH-ing into servers, finding processes, and manually killing them—a process that can take precious minutes while damage continues.
The deployment has no designed way to quickly stop the agent.
The agent runs across multiple processes or containers with no central control.
Containerization or systemd automatically restarts killed processes.
Queued or scheduled tasks continue executing even after main process stops.
Webhooks or integrations continue invoking the agent from outside.
During a security incident, a team discovered their AI agent was compromised at 2 AM:
1. The on-call engineer noticed unusual outbound traffic 2. They couldn't SSH in because the attacker had changed credentials 3. They tried to stop the cloud instance but the agent was in a serverless function 4. The function kept being invoked by a webhook they couldn't quickly disable 5. It took 45 minutes to fully stop the agent 6. By then, significant data had been exfiltrated
A one-click kill switch would have stopped the breach in seconds instead of minutes.
When you self-host your OpenClaw, you're responsible for addressing these risks:
Clawctl includes built-in protection against no kill switch:
Instantly terminate any agent with a single click from the dashboard. No SSH, no hunting for processes.
Kill switch stops all agent processes, cancels queued work, and pauses external triggers.
Trigger kill switch from your phone. Respond to incidents from anywhere.
Configure automatic shutdown based on anomaly detection or resource thresholds.
Shutdown preserves state for investigation. Understand what happened before resuming.
Whether you use Clawctl or not, follow these best practices:
Clawctl includes enterprise-grade protection against this threat and many others. Deploy your OpenClaw securely in 60 seconds.