← back to paper
arxiv: 2606.22873 · 2 revisions
SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning