Safety and Security of LLM Agents - Challenges and Future Directions

Explore the critical safety and security challenges facing Large Language Model (LLM) agents in this comprehensive 53-minute conference talk from RSA Conference. Delve into a detailed risk overview that includes a comprehensive threat taxonomy specifically designed for LLM agents, while discovering novel mitigation techniques to address emerging vulnerabilities. Learn about cutting-edge attack vectors that target LLM agents and examine corresponding defense strategies to counter these threats. Gain practical insights into effectively managing risks associated with LLM agent deployment and discover how to harness their capabilities safely in real-world applications. Master the implementation of least-privilege principles as a fundamental security approach for LLM agent systems. This session provides essential knowledge for security professionals, AI practitioners, and researchers working with or planning to deploy LLM agents in production environments.