Securing LLM-Powered Apps: Overcoming Security and Privacy Challenges

This conference talk from Devoxx explores the security and privacy challenges of integrating Large Language Models (LLMs) into applications. Learn about critical vulnerabilities like prompt injection, key leakage, and misuse of private customer data for model training. Discover how existing security issues in your systems can affect LLM behavior and outcomes. Brian Vermeer and Lize Raes provide a comprehensive overview of potential risks when implementing AI capabilities such as database access and automated purchasing, while offering practical strategies for ensuring data privacy compliance and implementing security best practices for LLM-powered applications. The 44-minute presentation equips developers with essential knowledge to build secure AI-enhanced systems while navigating legal restrictions and protecting user data.