CrowdStrike unveils a kill switch for rogue AI agents

CrowdStrike has a new way to stop an AI agent that goes off script. The company unveiled the AI Kill Switch on Tuesday, a feature of its Falcon Guardian platform, and says it is available now. It has not published a price or said where it is sold.

An AI agent is a program that uses a language model to carry out tasks on its own, such as reading files, calling services or sending messages. Unlike a chatbot, it acts, so a bug or a malicious instruction can have real consequences.

One action by an authorized administrator does three things: it terminates the running agent, applies a policy that keeps it from restarting on the affected device or group of devices, and cuts its access to the corporate network. Every step is logged for audit.

For now the cut happens at the AI gateway, the checkpoint that carries the agent's traffic to the models. An agent that does not go through it is, in principle, out of reach. Enforcement directly on the device, through the Falcon sensor, is coming "soon," with no date, and the release does not say which gateways it supports.

CEO and founder George Kurtz, speaking at Fal.Con Europe in Barcelona, framed the tool as "how you get the full benefit of AI": with a kill switch, he said, a company stays in control.

All of this comes from the vendor. There is no independent testing of how long the cut takes or how it performs against a real agent.

Falcon Guardian, which the Austin, Texas, company launched on Sept. 1, already discovers AI agents and enforces policies on them.

CrowdStrike also said Tuesday it will acquire Fiddler, which monitors agents, and Dreadnode, which tests AI systems against more than 100 known attacks, including jailbreaks and prompt injection. Both are due in the fourth quarter of CrowdStrike's fiscal 2027. Fiddler will be sold as Falcon Observe.