What it means
An agent kill switch is a safety mechanism designed to immediately stop an autonomous AI agent's operations. When an AI agent is working independently — making decisions, interacting with software, or managing data — it can sometimes behave in ways that are unexpected, harmful, or outside its intended goals. A kill switch gives a human operator or an automated monitoring system a way to intervene and freeze the agent's actions instantly. This prevents the AI from causing further damage, making costly errors, or violating security protocols. It acts as a digital emergency brake to regain control when the system's behavior becomes unpredictable or dangerous.
Why it matters for governance
It provides the critical layer of human oversight and risk mitigation that keeps autonomous systems controllable during unforeseen failures.
Example
A trading agent starts placing orders far above its normal volume; the kill switch freezes it mid-session, stopping any further orders before the account is drained.