CrowdStrike is betting that he has found a solution to deal with out-of-control artificial intelligence.
On Tuesday, this cybersecurity provider launched a “emergency stop” switch, designed to allow security teams to shut down out-of-control internal AI agents, prevent them from restarting, and cut off their access to the corporate network. This tool is built into the Falcon Guardian platform of CrowdStrike; the platform was launched in September with the aim of protecting AI agents. CrowdStrike also announced integration with Anthropic’s Claude Cowork.
CrowdStrike Chief Commercial Officer Daniel Bernard said to CNBC: "I believe that, as the AI application enters the next phase, it is indeed very important to have the ability to control, shut down, and make judgments and decisions."
A recent series of highly publicized cybersecurity incidents has made the call for the establishment of some kind of emergency shutdown mechanism even more urgent from the outside world.
Among these, the intrusion incident of Hugging Face planned by the out-of-control OpenAI entity in July has particularly raised concerns among companies. Last week, Binance disclosed the largest cryptocurrency cyberattack on record; recently, hackers also targeted hundreds of refrigerators and freezers in several French supermarkets.
This summer, a U.S. congressman proposed a bipartisan bill that requires AI companies to incorporate a shutdown capability into their models.
However, critics argue that implementing emergency stop buttons across multiple systems, data centers, and complex AI models is simply too complicated.
Bernard indicates that in the systems where customers use the CrowdStrike Falcon platform, the operation of this "emergency stop" switch is just like that of a light switch. He added that this process can detect internal agent activities, but it still requires human judgment to determine whether they are benign or malicious.
Bernard said, "If you terminate a process at its source, and we have such access at the root level... that is quite binary. You either shut down AI and the agent, or you don't shut them down, and then you report on it afterwards."
In the AI era, enterprises are rapidly adopting autonomous tools, yet they often lack the necessary security protections and safeguards.
CrowdStrike believes that its new tool will enable companies to 'stop' the improper behavior of AI intelligents immediately, while still leaving room for innovation.
The company also announced plans to acquire Fiddler AI, a startup in observability, as well as Dreadnode, a security testing company.
As security requirements continue to rise, CrowdStrike and its competitors are rapidly acquiring cybersecurity startups. The demand for AI's security tools has also driven a broader wave of cybersecurity transactions.
