Furina: Fragmented Uncertainty-Driven Refusal Instability Attack 文章

ArXiv CS.AI2026-05-27NEWSen作者: Tongxi Wu, Jian Zhang, Yang Gao
大语言模型
查看原文 →
关系图谱

Furina: Fragmented Uncertainty-Driven Refusal Instability Attack · 相关技术

相关技术

Multimodal Large Language Models (MLLMs)ODEPLALLMlanguage model多模态safetyalignmentStraight-Through EstimatorSAFENPUNATMECLMMlarge language modelsHISGrouped Memorization EvaluationEffort Metric AttentionCISARGANNTableTPU