Hacking and Securing LLM Applications - Understanding Browser Control Security Risks

Explore a comprehensive 40-minute conference talk from DeepSec 2024 that examines the security implications and potential risks of giving Large Language Models (LLMs) autonomous control over systems. Learn about the vulnerabilities in autonomous browser and software engineering agents, understand the mechanics of LLM agents, and discover how attackers can exploit these systems through indirect prompt injection. Through practical demonstrations, including attacks on browser agents and coding assistants, gain insights into data exfiltration techniques and the challenges of securing LLM applications. Master battle-tested security strategies for protecting LLM-powered systems, with access to additional resources including CTF challenges, research articles, and insights from leading LLM security experts. The presentation includes detailed demonstrations, real-world examples, and practical mitigation strategies for developers and security professionals working with LLM applications.