Definition
A technical control that lets an operator immediately stop a running AI agent — or a specific tool call — the moment it does something dangerous, even if the agent's own plumbing is compromised. Unlike a governance mandate requiring a human-off switch, this is the implemented mechanism, deployable at runtime and able to cover shadow (unregistered) agents as well as managed ones.
Why it matters
When an agent goes off-script, deception or accidents escalate in seconds; a working kill switch is what turns 'we have a policy' into 'we can actually stop it.'