The shift from conversational AI to agentic AI is forcing a reevaluation of enterprise security. Cisco's Jeetu Patel warned that AI agents, while highly capable, lack fear of consequences and could cause irreversible damage if not properly controlled.
Patel compared agents to 'teenagers,' emphasizing the need for action control rather than traditional access control. He stressed that every agent action must be observable, interceptable, and subject to dynamic guardrails.
Cisco launched DefenseClaw, an open-source security framework designed to protect AI agents. The tool integrates with Nvidia's OpenShell and activates security services automatically during agent operations.
Patel noted that the agentic AI security market is worth hundreds of billions, but the risks of poor implementation remain poorly understood.