A New Side Channel Attack on AI Assistants - Exploiting Vulnerabilities in Encrypted Messages

Explore a groundbreaking cybersecurity presentation from DEF CON 32 that exposes a critical side-channel vulnerability affecting major AI assistants. Discover how encrypted conversations with AI systems like ChatGPT could be intercepted and read by malicious actors, compromising sensitive user interactions ranging from personal health queries to professional communications. Learn about the technical intricacies of this novel exploit that impacted industry leaders including OpenAI, Microsoft, Cloudflare, Quora, and Notion. Gain practical insights into identifying similar vulnerabilities in AI assistants through network traffic analysis, understanding attack models, and comprehending the broader implications of this security discovery. Through real-world examples and detailed technical demonstrations, master the knowledge needed to recognize and understand these emerging security challenges in AI systems.