A single optimized image, trained through the vision and language modules, makes aligned multimodal LLMs produce dangerous responses across diverse prompts and some models.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Universal Adversarial Attack on Aligned Multimodal LLMs
A single optimized image, trained through the vision and language modules, makes aligned multimodal LLMs produce dangerous responses across diverse prompts and some models.