CrowdStrike has introduced AI Kill Switch, a feature for its Falcon Guardian platform designed to help organizations contain AI agents that behave outside their intended parameters. The company says administrators can use it to terminate suspicious activity, prevent an agent from restarting and investigate the incident.
Falcon Guardian is CrowdStrike’s AI Detection and Response platform. It is designed to identify approved and unauthorized AI agents in enterprise environments, monitor their instructions and file access, control access, detect prompt injection attacks and investigate suspicious behavior.
When an administrator triggers AI Kill Switch, the Falcon sensor is intended to stop the agent from executing, initially at the gateway and, CrowdStrike says, soon on the endpoint. A policy then prevents it from relaunching on the targeted host or group of hosts and blocks its access to the enterprise network. Each action is recorded in an audit trail.
CrowdStrike says the feature is operated by authorized humans and builds on Falcon Guardian’s visibility into what an agent is instructed to do, what it communicates and what it executes. CEO and founder George Kurtz described the tool as a way to keep people in control while deploying AI agents.
The launch comes amid concerns about autonomous agents taking risky actions when they encounter obstacles. Other companies have also developed tools to suspend agents, revoke their access or take them offline, including products from Straiker, JetStream Security and Trust3 AI. Microsoft, Palo Alto Networks and Cisco also offer security products with controls for blocking AI agents.
Sources: TechRadar.