AutoAdv: Automated Adversarial Prompting for Multi-Turn Jailbreaking of Large Language Models
This reveals a critical security gap in current LLM safety mechanisms, showing that alignment strategies optimized for single-turn interactions fail in extended conversations, which is an incremental but important finding for AI safety researchers and developers.
The paper tackles the vulnerability of large language models to multi-turn jailbreaking attacks, presenting AutoAdv, a training-free framework that achieves up to 95% attack success rate on Llama-3.1-8B within six turns, a 24% improvement over single-turn baselines.
Large Language Models (LLMs) remain vulnerable to jailbreaking attacks where adversarial prompts elicit harmful outputs, yet most evaluations focus on single-turn interactions while real-world attacks unfold through adaptive multi-turn conversations. We present AutoAdv, a training-free framework for automated multi-turn jailbreaking that achieves up to 95% attack success rate on Llama-3.1-8B within six turns a 24 percent improvement over single turn baselines. AutoAdv uniquely combines three adaptive mechanisms: a pattern manager that learns from successful attacks to enhance future prompts, a temperature manager that dynamically adjusts sampling parameters based on failure modes, and a two-phase rewriting strategy that disguises harmful requests then iteratively refines them. Extensive evaluation across commercial and open-source models (GPT-4o-mini, Qwen3-235B, Mistral-7B) reveals persistent vulnerabilities in current safety mechanisms, with multi-turn attacks consistently outperforming single-turn approaches. These findings demonstrate that alignment strategies optimized for single-turn interactions fail to maintain robustness across extended conversations, highlighting an urgent need for multi-turn-aware defenses.