# vision-language

Cross-Modal Embedding Attacks

Exploitation of shared embedding spaces across modalities: CLIP adversarial images, typographic attacks, VLM injection, and dimensionality reduction attacks.