一作ACL 2026Poster· Natural Language Processing
Mitigating Safety Context Amnesia in Multimodal Reasoning Models via Intent-Guided Safety Reasoning
Xiyao Dong · Guangsheng Cheng · YiLong Chen · Xiaojin Zhang · Kun He
摘要
Recent advances in Multimodal Large Reasoning Models (MLRMs) have enabled explicit chain-of-thought inference across vision and language, substantially improving performance on complex reasoning tasks. Despite these gains, the reasoning process introduces a subtle yet critical vulnerability. We i...