7 papers
CASA: Classification Augmented with Safety Attention for Robust Multimodal Alignment
Anurag Kumar, Raghuveer Peri, Jon Burnsky +4
Multimodal large-language models (MLLMs) often experience degraded safety alignment when harmful queries exploit cross-modal interactions. Models aligned on text alone show a highe…
A Closer Look at Adversarial Suffix Learning for Jailbreaking LLMs: Augmented Adversarial Trigger Learning
Zhe Wang, Yanjun Qi
Gradient optimization-based adversarial attack methods automate the learning of adversarial triggers to generate jailbreak prompts or leak system prompts. In this work, we take a c…
Graph of Attacks with Pruning: Optimizing Stealthy Jailbreak Prompt Generation for Enhanced LLM Content Moderation
Daniel Schwartz, Dmitriy Bespalov, Zhe Wang +2
As large language models (LLMs) become increasingly prevalent, ensuring their robustness against adversarial misuse is crucial. This paper introduces the GAP (Graph of Attacks with…
LaRS: Latent Reasoning Skills for Chain-of-Thought Reasoning
Zifan Xu, Haozhu Wang, Dmitriy Bespalov +3
Chain-of-thought (CoT) prompting is a popular in-context learning (ICL) approach for large language models (LLMs), especially when tackling complex reasoning tasks. Traditional ICL…
TurboFuzzLLM: Turbocharging Mutation-based Fuzzing for Effectively Jailbreaking Large Language Models in Practice
Aman Goel, Xian Carrie Wu, Zhe Wang +2
Jailbreaking large-language models (LLMs) involves testing their robustness against adversarial prompts and evaluating their ability to withstand prompt attacks that could elicit u…
TaeBench: Improving Quality of Toxic Adversarial Examples
Xuan Zhu, Dmitriy Bespalov, Liwen You +2
Toxicity text detectors can be vulnerable to adversarial examples - small perturbations to input text that fool the systems into wrong detection. Existing attack algorithms are tim…