As AI agents grow increasingly capable and autonomous, organizations face mounting security challenges in their deployment. Anthropic has introduced a Zero Trust security framework specifically designed for AI agents, addressing the unique risks these systems present. The framework applies traditional cybersecurity principles—long established in network security—to the emerging challenge of controlling and monitoring autonomous AI systems that make independent decisions. In a Practical AI episode, hosts Chris Benson and Daniel Whitenack examine how Zero Trust principles can be adapted for agentic AI systems. The discussion covers the core security risks that distinguish AI agents from traditional software, including the potential for agents to take unexpected autonomous actions and the difficulty of predicting agent behavior in novel situations. The hosts break down practical security controls and explore how organizations can implement Zero Trust architectures to safely deploy AI agents in high-impact environments. The conversation emphasizes that traditional cybersecurity best practices require evolution for the AI agent era. Organizations must move beyond assuming trust within their systems and instead implement continuous verification, least-privilege access, and comprehensive monitoring of agent behavior. This shift is particularly critical for enterprises deploying agents in sensitive domains where failures could have significant consequences.