VLMShield detects malicious multimodal prompts in VLMs as an efficient add-on by exploiting distributional differences in features aggregated from an extended CLIP model.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
VLMShield: Efficient and Robust Defense of Vision-Language Models against Malicious Prompts
VLMShield detects malicious multimodal prompts in VLMs as an efficient add-on by exploiting distributional differences in features aggregated from an extended CLIP model.