Securing AI Agents - Exploring Critical Threats and Exploitation Techniques

Explore critical security threats facing autonomous AI agents through comprehensive threat modeling, live hacking demonstrations, and advanced exploitation techniques in this 42-minute conference talk from Security BSides San Francisco 2025. Dive deep into real-world autonomous AI system vulnerabilities while examining model poisoning attacks through hands-on demonstrations that reveal how these systems can be compromised. Learn about sophisticated prompt injection techniques that can manipulate AI agent behavior and discover practical mitigation strategies to defend against these emerging threats. Gain insights into the unique security challenges that autonomous AI agents present compared to traditional AI systems, and understand how to implement robust security measures to protect these increasingly prevalent technologies in enterprise and consumer environments.