Page 147 - Cyber Defense eMagazine March 2024

How Large Language Models Can Be Used to

Weaponize AI

By Bobby Cornwell, Vice President Strategic Partner Enablement & Integration, SonicWall

In December, I predicted threat actors would tap into chat-based and generative artificial intelligence
(gen-AI) tools to augment and enhance attack methods and discover new attack vectors in the coming
year. Sadly, it appears this prediction has come true sooner than expected. This month we witnessed
what some say was the largest data leak in history (26 billion exposed records), and we saw that North
Korean hackers are starting to use gen-AI to conduct cyberattacks and search for targets to hack.

At the heart of this burgeoning cyberthreat are new, large language modules (LLMs) that enable more
efficient ways to compromise vast numbers of computing devices and launch increasingly more
successful cyberattacks, which, in turn, net attackers even more valuable data that can be fed back into
their expanding data lakes.

LLMs are able to bring together massive amounts of breach data that have been cobbled together from
numerous historical breaches to train new AI-based systems. This novel approach gives aspiring hackers
access to volumes of personal data that includes PII (personal identifiable information) and login
credentials to any number of websites that have historically been compromised.

Cyber Defense eMagazine – March 2024 Edition 147
Copyright © 2024, Cyber Defense Magazine. All rights reserved worldwide.