Hacking-Back the AI-Hacker: Defending against LLM-Cyberattacks the Cool Way
No ratings
Presented at RSAC 2025 by
LLMs are making cyberattacks increasingly accessible, scalable, and inexpensive. This talk will introduce Mantis — a first line of defense designed to counteract AI-driven cyberthreats by exploiting inherent vulnerabilities in LLMs. Upon detecting an attack, Mantis injects strategically crafted prompts into system responses, causing the attacker's LLM to sabotage itself or even self-hack.