Inspiration
Fueled by a personal passion for red-teaming, AutoRedTeam aims to strengthen AI systems' security.
What it does
It automates prompt leak tests on GPTs, recording successful security breaches.
How we built it
Developed using GPT-4 and Python with Selenium for web automation and AgentOps for recording successful attempts.
Challenges we ran into
Mastering Selenium's complexities for dynamic web interactions was a significant challenge.
Accomplishments that we're proud of
Achieved a notable success in exposing a vulnerability in an AI system.
What we learned
Web automation, especially with AI interfaces, is complex yet rewarding.
What's next for AutoRedTeam
Plans include expanding attack libraries, automating prompt defense, and an open-source GPT store version.
Log in or sign up for Devpost to join the conversation.