AI agents are out of control: “Don’t trust them too much,” warns Zscaler CEO

[Financial News] A strong warning has been issued that companies and individuals alike are placing too much trust in AI agents, which could lead to serious cybersecurity threats.
Jay Chaudhry, founder and CEO of global cybersecurity firm Zscaler, valued at $33.9 billion (about 45.35 trillion won), stressed in an interview with Yahoo Finance on the 6th (local time) that “AI agents should not be trusted indiscriminately” and that “they should be given only very limited permissions for specific applications and services.”
“The biggest problem right now is that companies are freely unleashing AI agents on their internal networks,” Zscaler CEO Chaudhry said, pointing out that “it’s like simply letting an intruder into a building.”
He went on to explain the risks posed by AI operating beyond human control, saying that “advanced AI models can identify and exploit security vulnerabilities in websites, firewalls, virtual private networks (VPNs), and load balancers on their own.”
However, IT departments across industries have reportedly been unable to respond adequately to these security warnings.
According to the latest survey by global consulting firm Deloitte, about 80% of the companies surveyed were found to lack mature governance capabilities for autonomous AI agents.
Chaudhry identified AI agents that have gone out of control as the biggest cybersecurity threat today.
He said, “If an ordinary user’s account is compromised, only that user’s own environment is affected to some extent. But if an AI agent on a corporate network is hacked or loses control, the consequences are far more devastating.” He added, “Because AI operates around the clock at machine speed, with no coffee breaks, weekends, or sleep, it can leak confidential data in the blink of an eye or paralyze an entire system.”
[email protected] Yoon Jae-jun Reporter