Recently, I’ve been cooking up some analysis of AI red teaming attacks, sourced from Crucible challenges at Dreadnode. Excited to share our new paper, focusing on how automation can amplify offensive AI techniques:
arxiv.org/abs/2504.19855
arxiv.org
The Automation Advantage in AI Red Teaming
This paper analyzes Large Language Model (LLM) security vulnerabilities based on data from Crucible, encompassing 214,271 attack attempts by 1,674 users across 30 LLM challenges. Our findings reveal a...
