Hacc-Man: An Arcade Game for Jailbreaking LLMs
Matheus Valentim, Jeanette Falk, Nanna Inie
TL;DR
The paper presents Hacc-Man, an arcade-style game designed to explore jailbreaking of LLMs, positioned at the crossroads of security and creativity research. It articulates three contributions: raising awareness of jailbreak risks, increasing players' self-efficacy in interacting with LLMs, and collecting data on creative problem solving strategies used to jailbreak models. The work surveys related literature on jailbreaking, red teaming, self-efficacy, and creativity, and introduces a research-through-design approach with six diverse challenges and a data-collection framework. The open-access demo and dataset aim to foster further research on LLM weaknesses, defense strategies, and the cognitive processes underlying promptcrafting and problem solving in adversarial settings.
Abstract
The recent leaps in complexity and fluency of Large Language Models (LLMs) mean that, for the first time in human history, people can interact with computers using natural language alone. This creates monumental possibilities of automation and accessibility of computing, but also raises severe security and safety threats: When everyone can interact with LLMs, everyone can potentially break into the systems running LLMs. All it takes is creative use of language. This paper presents Hacc-Man, a game which challenges its players to "jailbreak" an LLM: subvert the LLM to output something that it is not intended to. Jailbreaking is at the intersection between creative problem solving and LLM security. The purpose of the game is threefold: 1. To heighten awareness of the risks of deploying fragile LLMs in everyday systems, 2. To heighten people's self-efficacy in interacting with LLMs, and 3. To discover the creative problem solving strategies, people deploy in this novel context.