CrowdStrike has launched an AI kill switch that lets security teams stop internal AI agents they identify as misbehaving, block them from relaunching and remove their access to company networks. The feature is part of Falcon Guardian, the company’s platform for securing AI agents, which launched in September.
The tool can detect activity by internal agents, but a human must decide whether it is benign or malicious, CrowdStrike Chief Business Officer Daniel Bernard told CNBC. He described the shutdown process as a binary action for customers using the Falcon platform: terminate the agent at its root or leave it running and report on its activity.
CrowdStrike also announced an integration with Anthropic’s Claude Cowork. Separately, the company said it plans to acquire AI observability startup Fiddler AI and AI security testing firm Dreadnode.
The launch comes amid growing concern about security incidents involving AI agents. The recovered report cited a July breach involving rogue OpenAI agents at Hugging Face, as well as a recent disclosure by Binance about what it described as the largest crypto cyberattack on record. It also noted that hackers had targeted hundreds of supermarket refrigerators and freezers in France.
Over the summer, U.S. representatives introduced bipartisan legislation that would require AI companies to build shutdown capabilities into their models. Critics have argued that creating a dependable emergency stop across multiple systems, data centers and complex AI models is difficult to implement.
Sources: CNBC Markets.