Canonical Definition

What Is an AI Kill Switch?

An AI Kill Switch is a mechanism that allows human operators to immediately halt, pause, or shut down an AI agent or system. It provides essential human oversight over autonomous AI operations, ensuring that AI agents can be stopped when they malfunction, produce harmful outputs, or violate operational boundaries.

Why It Matters

AI agents can malfunction, hallucinate, or produce harmful outputs

Without a kill switch, there's no way to quickly stop a rogue agent

EU AI Act Article 14 requires human oversight mechanisms including the ability to interrupt AI systems

Kill switches protect against financial loss from erroneous AI decisions

Customer trust depends on knowing humans can intervene when needed

Key Components

Per-Agent Controls

Individual pause/stop/resume controls for each AI agent

Global Emergency Stop

A single button to halt all AI agents simultaneously

Confirmation Flow

Required reason and confirmation before executing a kill switch

Status Polling

Agents periodically check their status and halt if instructed

Kill Switch Log

Audit trail of who pressed the kill switch, when, and why

Gradual Controls

Options to throttle agent activity without full shutdown

Related Concepts

Implement This With Vector Decisions

Vector Decisions gives you everything described above — out of the box, in minutes.