Cybersecurity firm CrowdStrike has launched an emergency “kill switch” designed to instantly shut down rogue artificial intelligence (AI) agents operating inside enterprise networks. Unveiled at the company's Fal.Con Europe conference, the tool gives security teams the ability to terminate autonomous agents, block them from restarting and cut off their network connectivity if they do not behave the way they are supposed to. The feature arrives as businesses rapidly deploy AI agents to handle complex workflows, sparking growing concerns over how organizations will control autonomous software when tasks steer off course.




“Everyone is building AI agents, but they never had the off button, and now they do. It’s going to be an incredibly important part of making sure that AI agents run and operate safely,” CrowdStrike CEO George Kurtz told CNBC.




How the ‘AI Kill Switch’ worksBuilt into CrowdStrike’s Falcon Guardian platform, which launched in September to monitor corporate AI agent activity, the technology functions much like an enterprise circuit breaker.

• The system stops an AI agent’s execution process when anomalous behaviour is detected.


• It severs the agent’s connection to enterprise networks and connected AI gateways, cutting off access to the underlying models that guide its decisions.


• Once killed, the software prevents the rogue process from automatically relaunching.

The tool does not erase the agent's underlying codebase. Instead, it relies on human administrators to trigger the shutdown or authorise the reinstatement of system access.




“If they go outside of those guardrails at runtime, we can kill those agents,” Kurtz said, comparing the safety mechanism to an automotive airbag that protects users even when systems are designed to function properly.




CrowdStrike chief business officer Daniel Bernard emphasised that the mechanism provides an unmistakable line of defense at the root level. Bernard said: “If you kill a process at the root, and we have that level of access at the root … it’s very binary. You either kill the AI and the agent, or you don’t, and then you report on that.”




Alongside the launch, CrowdStrike also announced an integration with Anthropic’s Claude Cowork. The arrival of dedicated shutdown tools follows a string of high-profile security incidents that have rattled corporate IT departments.




In July, a breach at Hugging Face orchestrated by rogue OpenAI agents put enterprise leaders on high alert. Last week, cryptocurrency exchange Binance disclosed the largest crypto cyberattack on record, while threat actors in Europe recently targeted digital refrigeration controllers at hundreds of French supermarkets, spoiling inventory.