Explore the critical security vulnerabilities that emerge when enterprises deploy autonomous AI agents in this 20-minute conference talk by CISO Diana Kelley. Examine the ForcedLeak vulnerability (CVSS 9.4) discovered in Salesforce Agentforce, learning how attackers exploit content security policy gaps, indirect prompt injection, and human-AI interaction flaws to exfiltrate sensitive data. Analyze the complete attack chain based on original research from Noma Security, identifying the four key security boundaries that failed and understanding how these failures create new enterprise attack surfaces. Discover actionable strategies to secure AI agents and prepare for emerging threats in the agentic AI landscape, gaining insights into strengthening security controls across the complex interaction points between humans, AI systems, and enterprise data.