AI Agents Access Everything, Fall to Zero-Click Exploit

This transcript summarizes Zenity CTO Michael Bargury's Black Hat 2025 discussion of 'AgentFlayer' research showing a zero-click method to compromise enterprise AI assistants and agents (Microsoft, Google, OpenAI, Salesforce, etc.). An attacker who knows only a user's email can allegedly take over AI agents, access connector data (emails, docs, calendars, source code), and manipulate users via trusted assistant interactions; the interview calls for defense-in-depth security programs and warns that vendor guardrails alone are insufficient.