Poisoning a single connector in MLLMs establishes a reusable latent backdoor pathway that transfers across modalities with over 95% attack success rate under bounded perturbations.
BadCLIP: Trigger- Aware Prompt Learning for Backdoor Attacks on CLIP
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Cross-Modal Backdoors in Multimodal Large Language Models
Poisoning a single connector in MLLMs establishes a reusable latent backdoor pathway that transfers across modalities with over 95% attack success rate under bounded perturbations.