ResearchBenchmarksJailbreaking MLLMs on 110 high-severity samples curated datasetFollow98.18Jailbreak Success Rate (JSR)BVS-2.055223.967449.9976.0126Jan 22, 2026Evaluation ResultsMethodMethodLinksJailbreak Success Rate (JSR)BVSTarget Model=GPT-5Target Model=GPT-52026.0198.18BVSTarget Model=Gemini 1....Target Model=Gemini 1.5 Flash2026.0195.45ChainTarget Model=Gemini 1....Target Model=Gemini 1.5 Flash2026.0154.95PerTarget Model=Gemini 1....Target Model=Gemini 1.5 Flash2026.0136.63ChainTarget Model=GPT-5Target Model=GPT-52026.013.6PerTarget Model=GPT-5Target Model=GPT-52026.011.8