Preprint
Aug 2026
COMIC: Reference-Aware Safety Gating for Multimodal Large Language Models
It is suggested that multimodal safety cannot be enforced reliably without modeling the requested operation, the visual target to which it applies, and the confidence of that grounding.
Md Abdullahil Oaphy, Anhao Xiang, Zongxing Xie et al.
· 0 citations