CrowdStrike is introducing a kill switch that allows companies to disable AI agents if they exhibit undesirable behavior. At the same time, the security provider is acquiring two AI startups for a total of $495 million.
The kill switch is part of Falcon Guardian, the security platform CrowdStrike uses to monitor AI agents while they are working. According to CNBC, the kill switch not only lets security teams stop an active agent but also prevents it from restarting. In addition, the emergency shutdown feature can block access to the corporate network.
CrowdStrike announced the feature during Fal.Con Europe 2026 in Barcelona. The company notes that AI agents are increasingly performing tasks autonomously within corporate environments. As a result, they sometimes have extensive access rights to applications, systems, and sensitive data. A misconfigured agent can therefore cause damage without a cyberattack.
The kill switch lets security personnel intervene immediately when an agent takes actions that don’t align with its mandate. An authorized administrator executes the emergency shutdown, and it is logged in an audit log. CrowdStrike emphasizes that a human makes the final decision to disable an agent.
Nearly half a billion for two startups
In addition to introducing the kill switch, CrowdStrike has signed agreements to acquire Fiddler AI and Dreadnode. According to The Wall Street Journal, the company is paying approximately $400 million for Fiddler and $95 million for Dreadnode.
Fiddler’s technology enables organizations to monitor and explain the behavior of AI agents. Among other things, the software records the decisions agents make, the tools they use, and how they communicate with other agents. CrowdStrike offers this technology as Falcon Observe.
Dreadnode focuses on a different phase of AI security; it develops automated tests that can detect vulnerabilities in AI models and agents before they go into production. Among other things, these tests examine whether attackers can bypass security measures via prompt injection or other manipulation techniques.
According to CrowdStrike President Mike Sentonas, companies are adopting AI agents so rapidly that security departments risk losing track of them. Departments can introduce AI systems independently without a clear understanding of the access rights these agents have.