← All Terms

Circuit Breaker

An automatic control that halts an AI agent or stops its actions spreading when a threshold or anomaly is detected, pending human review.

Agentic AI

A circuit breaker for AI agents borrows the idea from electrical systems and software engineering: when something goes wrong, cut the circuit before the damage spreads. It trips on a defined condition, such as a spend cap reached, a decoy record touched or a burst of denied requests, and freezes the agent until a person releases it.

Singapore’s Cyber Security Agency calls for circuit breakers “that freeze propagation when anomalous behaviour is detected” in its 2026 addendum on securing agentic AI. They matter most in multi-agent systems, where one hijacked agent can pass bad instructions to others within a few turns.

A circuit breaker differs from a kill switch in who pulls it. The kill switch is a human decision to stop a system; the circuit breaker acts on its own, inside the seconds an agent works in, and hands the decision to a person afterwards.

Read more on this
Security

Put Boundaries Around Autonomy: Practical Controls for Rogue Agent Risk

If an AI agent can always be fooled some of the time, the controls that matter limit what it can do when it is. Maria Singh and I set out five autonomy tiers, the hard limits that hold outside the model, human approval that is more than a rubber stamp, kill switches that have been tested, and a 90-day plan for banks and government agencies. Part 3 of 3 on rogue AI agents.

Security

Follow the Actions: Detecting Rogue AI Agents Before the Damage Spreads

A rogue AI agent says reasonable things while it does unreasonable ones. The first AI-run espionage campaign was caught by its tempo and volume, not its words. Maria Singh and I set out twelve action signals a security team can alert on, how damage spreads between agents, and the logs and clocks regulators now expect. Part 2 of 3 on rogue AI agents.