COMPEL Glossary / kill-switch
Kill Switch
A kill switch is an immediate, unconditional mechanism to halt an AI agent's operation.
What this means in practice
Every agent deployed under the COMPEL framework must have a kill switch, with design requirements escalating based on autonomy level and risk tier. For low-autonomy agents, a manual kill switch accessible to the operational owner is sufficient. For Level 3 agents, both manual and automated kill switches are required, with automated triggers tied to behavioral boundaries. For Level 4 agents, kill switches must operate with sub-second response times, be independent of the agent's own infrastructure, and be tested on a defined schedule. For critical-risk agents at any autonomy level, kill switches must be independently operable by at least two designated personnel and trigger automatic notification to the governance function.
Why it matters
Every AI agent must have a reliable mechanism for immediate, unconditional shutdown. Without kill switches, organizations cannot contain AI systems that exhibit unexpected behavior, exceed authorized boundaries, or cause harm. The consequences of an ungovernable agent can escalate rapidly, especially in systems with financial transaction authority, data modification capabilities, or external communication channels.
How COMPEL uses it
Kill switch requirements escalate based on autonomy level and risk tier in COMPEL's Agent Governance framework. Low-autonomy agents require manual kill switches. Level 3 agents need both manual and automated switches with behavioral boundary triggers. Level 4 agents require sub-second response times independent of the agent's own infrastructure. Critical-risk agents at any level require dual-person operation and automatic governance notification.
Related Terms
Other glossary terms mentioned in this entry's definition and context.