LLaVAShield: Safeguarding Multimodal Multi-Turn Dialogues in Vision-Language Models Paper • 2509.25896 • Published Sep 30, 2025 • 1