As AI safety becomes a focus issue, CrowdStrike (CRWD.US) launches AI “emergency stop switch”

The Zhitong Finance App learned that CrowdStrike (CRWD.US) is betting that it has a solution to deal with uncontrolled artificial intelligence. The cybersecurity vendor launched an AI “AI kill switch” (AI kill switch) on Tuesday to allow security teams to stop AI agents that “deviate” within the enterprise, block the way they can be restarted, and cut off their access to the corporate network. The tool is built into the Falcon Guardian platform launched by CrowdStrike in September to secure AI agents. CrowdStrike also announced an integration with Anthropic's Claude Cowork.

“Having the ability to control and shut down, and being able to make decisions and decide (whether to shut down), I think this is critical to the next stage of AI applications,” said Daniel Bernard, Chief Commercial Officer at CrowdStrike.

Recently, a series of high-profile cyber attacks have added a sense of urgency to calls for the establishment of some kind of emergency shutdown mechanism. The July Hugging Face intrusion, orchestrated by an uncontrolled OpenAI agent, alerted the business community in particular. Last week, Binance revealed the largest cryptocurrency cyber attack on record; hackers also recently attacked hundreds of refrigerators and freezers in French supermarkets.

This summer, US congressmen proposed a bipartisan bill requiring AI companies to build shutdown capabilities into the model. Critics believe, however, that implementing an emergency stop button across multiple systems, data centers, and complex AI models is far more complicated than reality.

Bernard said that in systems where customers use their Falcon platform, the shutdown switch works like a circuit breaker. The process can detect the activity of internal agents, but it is up to humans to decide whether to cut off an agent and restore access to it.

“If you kill a process at the root, and we have exactly this level of access at the root... then it's very dualistic: you either kill the AI and the agent, or you don't, and then you report it,” Bernard said.

CrowdStrike said the shutdown switch does not remove the agent's code, but rather cuts off its access to AI models that influence decisions through connected AI gateways.

In the AI era, enterprises are rapidly adopting autonomous tools, yet they often lack necessary security and protection measures. CrowdStrike believes this new tool will allow companies to kill the “misconduct” of AI agents in the bud while leaving room for innovation.

The company also announced plans to acquire AI observability startup Fiddler AI and AI security testing company Dreadnode.

CrowdStrike and its rivals are racing to acquire cybersecurity startups to keep up with growing security demands. The need for AI security tools is also driving a wider wave of mergers and acquisitions in the cybersecurity industry.