How NOT to Train Your Hack Bot: Dos and Don'ts of Building Offensive GPTs

No ratings

Presented at BlackHat USA 2023 by

Shane Caldwell

Ariel Herbert-Voss

No doubt everybody is curious if you can use large language models (LLMs) for offensive security operations. In this talk, we will demonstrate how you can and can't use LLMs like GPT4 to find security vulnerabilities in applications, and discuss in detail the promise and limitations of using LLMs this way. We will go deep on how LLMs work and share state-of-the-art techniques for using them in offensive contexts.