Security Tensors as a Cross-Modal Bridge: Extending Text-Aligned Safety to Vision in LVLM

Open in new window