Innovator Coffee EP-30 AI Agent Security: Rethinking Enterprise Defense in the Autonomous Era
Welcome to the Innovator Coffee, a podcast that bridges the gap between people and the world of AI and innovation. We're Tom and Wickey. Follow us to explore the top AI products, ecosystem insights, and the emerging trends. In this episode, we explore a rapidly emerging enterprise challenge: AI Agent Security. As AI agents are integrated into core systems like Slack, Jira, and CRM platforms, they are no longer just “tools.” They become autonomous digital entities with permissions, access, and the ability to take action. The problem? Traditional security tools were never designed to truly see or control them. In this conversation, we discuss: * Why “adding permissions + sandboxing + human approval” does not equal security * What “agent jacking” really means * How attacks can unfold through chains of actions that appear completely legitimate * How organizations can build sustainable security frameworks without slowing innovation If you are a CISO, founder, AI builder, or actively deploying agents into production environments, this episode is essential listening. Guest: Barak Sternberg, CEO& Co-Founder @ Stealth AI Security, Hacktivity Speaker, Ex-YL Ventures Advisor, Ex-Unit 8200 Timeline: 00:00:00 — Opening: Why AI Agent Security Is the New Battlefield 00:04:40 — The Core Problem: Agents Are Invisible to Traditional Security 00:11:50 — Identity Is a Baseline — But Not Enough 00:19:30 — The CISO’s Dilemma: Inventory Isn’t Control 00:24:50 — RPA vs. Agents: How “Agent Jacking” Happens 00:34:40 — The Sandbox Myth & The Illusion of Human Approval 00:43:30 — Market Landscape: Governance, Perimeter, Identity, and Agent TDR 00:51:30 — Closing: Why Observability ≠ Security — and What’s Missing Next Special thanks to Hannah Wang who did the wonderful job to assist to complete this podcast. https://www.linkedin.com/in/hannah-wang-9302421b3/ Hosts: Tom Kong *Stanford EE alumni, *Founder@ Stanford AGI Adventist Community (10K+ members so far from top VC, Engineers, startups from Silicon Valley ) *AI Lecturer, a serial entrepreneur in media and data. Advisor @ techtimes.com *AI deployment for 8 years, with NLP and recent LLMs (RAG, Agent, Diffusion) Wickey Wang *IT Security Compliance Leader & University Faculty *Growth fund VC advisor, VC fellow and Angel Investor with cybersecurity and AI focus *GAI Security book co-author Questions, Suggestions, Feedback and Comments? You can find us in LinkedIn: https://www.linkedin.com/in/wickey-wang-cisa-six-sigma-green-belt-2aaa913 https://www.linkedin.com/in/thomaskong/
Welcome to the Innovator Coffee, a podcast that bridges the gap between people and the world of AI and innovation. We're Tom and Wickey. Follow us to explore the top AI products, ecosystem insights, and the emerging trends. In this episode, we explore a rapidly emerging enterprise challenge: AI Agent Security. As AI agents are integrated into core systems like Slack, Jira, and CRM platforms, they are no longer just “tools.” They become autonomous digital entities with permissions, access, and the ability to take action. The problem? Traditional security tools were never designed to truly see or control them. In this conversation, we discuss: * Why “adding permissions + sandboxing + human approval” does not equal security * What “agent jacking” really means * How attacks can unfold through chains of actions that appear completely legitimate * How organizations can build sustainable security frameworks without slowing innovation If you are a CISO, founder, AI builder, or actively deploying agents into production environments, this episode is essential listening. Guest: Barak Sternberg, CEO& Co-Founder @ Stealth AI Security, Hacktivity Speaker, Ex-YL Ventures Advisor, Ex-Unit 8200 Timeline: 00:00:00 — Opening: Why AI Agent Security Is the New Battlefield 00:04:40 — The Core Problem: Agents Are Invisible to Traditional Security 00:11:50 — Identity Is a Baseline — But Not Enough 00:19:30 — The CISO’s Dilemma: Inventory Isn’t Control 00:24:50 — RPA vs. Agents: How “Agent Jacking” Happens 00:34:40 — The Sandbox Myth & The Illusion of Human Approval 00:43:30 — Market Landscape: Governance, Perimeter, Identity, and Agent TDR 00:51:30 — Closing: Why Observability ≠ Security — and What’s Missing Next Special thanks to Hannah Wang who did the wonderful job to assist to complete this podcast. https://www.linkedin.com/in/hannah-wang-9302421b3/ Hosts: Tom Kong *Stanford EE alumni, *Founder@ Stanford AGI Adventist Community (10K+ members so far from top VC, Engineers, startups from Silicon Valley ) *AI Lecturer, a serial entrepreneur in media and data. Advisor @ techtimes.com *AI deployment for 8 years, with NLP and recent LLMs (RAG, Agent, Diffusion) Wickey Wang *IT Security Compliance Leader & University Faculty *Growth fund VC advisor, VC fellow and Angel Investor with cybersecurity and AI focus *GAI Security book co-author Questions, Suggestions, Feedback and Comments? You can find us in LinkedIn: https://www.linkedin.com/in/wickey-wang-cisa-six-sigma-green-belt-2aaa913 https://www.linkedin.com/in/thomaskong/