LLM Proactive Defense Against Cyber Attacks
An interesting study: “Hacking the AI hacker: Rapid injection as a defense against LLM-driven cyberattacks“:
Large language models (LLMs) are increasingly being used to automate cyberattacks, making complex exploits more accessible and scalable. In response, we propose a new defense strategy designed to combat LLM-driven cyberattacks. We present Mantis, a defense framework that exploits LLM’s competitive input vulnerability to undermine malicious operations. When detecting an automated cyberattack, Mantis carefully injects input into the system’s responses, forcing the attacker’s LLM to disrupt its own operations (passive defense) or even compromise the attacker’s machine (active defense). By deploying targeted vulnerable decoy services to attract the attacker and using dynamic fast injections for the attacker’s LLM, Mantis can autonomously hack the attacker. In our experiments, Mantis consistently achieved more than 95% efficiency against LLM-driven automated attacks. For further research and collaboration, Mantis is available as an open source tool: this https url.
Of course, this is not a solution. But such things can be part of the solution.
Bruce Schneier sidebar photo by Joe McInnis.