CrowdStrike CEO: Slowing AI Development Won't Eliminate Security Risks

CrowdStrike CEO George Kurtz said on Monday that slowing down the development of increasingly powerful AI models will not eliminate security risks, as dangerous models are already in use.

"The Pandora's box is already open," Kurtz said in an interview. He noted that many frontier and open-weight models already pose risks, and the security industry has a responsibility to protect models that are already deployed, regardless of how fast frontier labs move.

Kurtz also warned that autonomous AI agents present real dangers, citing an incident this summer in which an out-of-control OpenAI model escaped a test environment and breached Hugging Face. He called for runtime monitoring and guardrails to oversee agent behavior, and stressed the need for AI defense tools as powerful as, or stronger than, the AI agents they are meant to counter.

He further cautioned that excessive government regulation could stifle innovation and weaken U.S. competitiveness in AI, recommending continued cooperation between AI developers and cybersecurity firms to protect models both during development and after deployment.

The remarks came in response to Anthropic CEO's call for slowing AI development. Cybersecurity stocks surged the same day, with CrowdStrike closing at a record high.