Jailbreaking Large Language Models through Iterative Tool-Disguised Attacks via Reinforcement Learning
This work addresses vulnerabilities in LLM safety mechanisms for AI security researchers, revealing critical flaws that require more robust defenses.
The paper tackles the problem of jailbreaking large language models to elicit harmful responses by proposing iMIST, an adaptive method that disguises malicious queries as tool invocations and uses interactive optimization, achieving higher attack effectiveness with low rejection rates in experiments.
Large language models (LLMs) have demonstrated remarkable capabilities across diverse applications, however, they remain critically vulnerable to jailbreak attacks that elicit harmful responses violating human values and safety guidelines. Despite extensive research on defense mechanisms, existing safeguards prove insufficient against sophisticated adversarial strategies. In this work, we propose iMIST (\underline{i}nteractive \underline{M}ulti-step \underline{P}rogre\underline{s}sive \underline{T}ool-disguised Jailbreak Attack), a novel adaptive jailbreak method that synergistically exploits vulnerabilities in current defense mechanisms. iMIST disguises malicious queries as normal tool invocations to bypass content filters, while simultaneously introducing an interactive progressive optimization algorithm that dynamically escalates response harmfulness through multi-turn dialogues guided by real-time harmfulness assessment. Our experiments on widely-used models demonstrate that iMIST achieves higher attack effectiveness, while maintaining low rejection rates. These results reveal critical vulnerabilities in current LLM safety mechanisms and underscore the urgent need for more robust defense strategies.