Abstract
GenAI CTF
AI Security CTF: Exploring Conversational Agent Vulnerabilities
In the AI Security CTF, participants delve into two critical vulnerabilities: Context Manipulation and Prompt Injection. The Finch's Terminal challenge reveals how context overload can inadvertently leak sensitive system details, while the Project Nightingale challenge demonstrates how role-play prompts can coerce AI into disclosing protected information. These scenarios underscore the importance of input validation, persona consistency, and context window management. By mastering these exploits, participants gain essential skills to fortify AI systems against evolving security threats.
Speaker
Timing
Starts at Saturday March 29 2025, 12:00 PM. The sessions runs for about 11 hours.