Compromising LLMs - The Advent of AI Malware

Explore the critical security implications of prompt injections in large language models like ChatGPT in this 36-minute Black Hat conference talk. Discover how a new generation of malware and manipulation techniques can operate entirely within AI systems, posing significant risks as companies rapidly integrate these models into various applications. Learn about potential vulnerabilities in future personal assistants and the consequences of compromised AI systems. Gain insights from experts Sahar Abdelnabi, Christoph Endres, Mario Fritz, Kai Greshake, and Shailesh Mishra on the urgent need for thorough security considerations in emerging AI technologies.