This report details a sophisticated black-box attack framework that utilizes fuzz testing to automatically craft concise and semantically coherent prompts capable of bypassing safety mechanisms in large language models (LLMs). The attack initiates with an empty seed pool and employs LLM-assisted mutation strategies, including Role-play, Contextualization, and Expansion, alongside a dual-level judge module to efficiently identify successful jailbreak attempts. The framework's efficacy has been demonstrated across various open-source and proprietary LLMs, achieving a success rate that surpasses existing benchmarks by over 60% in certain instances. Mitigation steps: **For AI Developers:** * Implement robust and multi-layered safety mechanisms that extend beyond simple input filtering and output restrictions. * Develop sophisticated detection methods that are resistant to semantically coherent adversarial prompts, utilizing contextual understanding and advanced anomaly detection techniques. **For Model Trainers/Fine-tuners:** * Regularly update and refine LLM safety models and datasets to adapt to evolving jailbreaking techniques. * Investigate techniques to detect and mitigate attacks based on prompt length and perplexity. * Employ adversarial training to enhance LLM robustness against diverse attack vectors.