LogiBreak: Logical Expressions Bypass LLM Safety Restrictions

Researchers have introduced LogiBreak, a novel black-box jailbreak method that converts harmful natural language prompts into formal logical expressions to circumvent safety systems in large language models (LLMs). The method exploits distributional discrepancies between alignment-oriented prompts and logic-based inputs, preserving semantic intent and readability while evading safety constraints. LogiBreak was evaluated on a multilingual jailbreak dataset spanning three languages, demonstrating effectiveness across various evaluation settings and linguistic contexts. The research is published on arXiv under computer science category.