Multi-Turn Adaptive Prompting Attack on Large Vision-Language Models

ArXi:2602.14399v2 Announce Type: replace Multi-turn jailbreak attacks have proven effective against text-only large language models (LLMs), where malicious content is gradually