AI Dictionary › Fondamenti AI
Kill Switch AI
An AI kill switch is a fast, reliable interruption mechanism by which a human operator can stop the execution of an AI system, particularly an autonomous agent taking actions in the real world, when its behavior turns out dangerous, unexpected, or out of control. Conceptually it is the last line of defense when all other preventive controls have failed.
At a technical level, an effective kill switch must be independent of the system it intends to stop: if the shutdown mechanism depends on the very process it should interrupt, a serious malfunction could compromise it along with everything else. That is why robust implementations separate the control channel from the agent's operating system, for example through an external supervising process with its own privileges, able to revoke access tokens, terminate processes, or cut network connectivity regardless of what the agent is doing at that moment.
It is a central topic in the design of AI agents with access to real tools, code execution, financial transactions, control of physical devices, where anomalous behavior not corrected in time can translate into concrete harm rather than just a wrong text reply. Many multi-agent orchestration platforms now include emergency stop controls as a design requirement from the outset.
The term "kill switch" originates in industrial engineering and mechanical systems, where it denotes a physical emergency switch to stop machinery; it was picked up in the AI safety debate, particularly in relation to autonomous agents, starting in the mid-2010s.
From our network
AGORÀ Intelligence: Enterprise AI Governance Platform
Govern AI at scale: policies, adoption and measurable results on your data. Built for boards and C-suite.
Visit agora-intelligence.com →From the Agora Intelligence blog
📱 Download the Android app (beta) iOS coming soon
Say what you mean. Get what you need.
Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.