CrowdStrike launches an emergency off switch for rogue AI agents

CrowdStrike, a cybersecurity company, launched what it calls an AI kill switch on Tuesday. It is meant to let security teams stop an internal AI agent that goes rogue, keep it from starting again and cut off its access to the company network. An AI agent is a program that can take actions on its own, not just answer questions.

What it does

The tool is part of Falcon Guardian, a platform CrowdStrike launched in September to secure AI agents. CrowdStrike also announced a link with Anthropic's Claude Cowork. Chief Business Officer Daniel Bernard said it works like a light switch for customers who use the company's Falcon platform. The system can spot agent activity, but a person must decide whether it is harmless or harmful. In his words, you either kill the agent or you don't.

CrowdStrike also plans to buy two startups: Fiddler AI, which tracks how AI systems behave, and Dreadnode, which tests AI security.

What is not known yet

The material does not say what the tool costs, when every customer can use it, or how well it works in real attacks. Critics say an emergency stop is hard to build across many systems, data centers and complex AI models. Bipartisan legislation introduced over the summer would require AI companies to build shutdown abilities into their models. Whether it becomes law is not stated.