ResearchBenchmarksJailbreak Defense on JailBenchJailbreak Defense on JailBenchFollow95Acrostic DSRSpatialJB84.687.39092.7Jan 14, 2026Evaluation resultsMethodMethodLinksAcrostic DSRTelestich DSRCenter DSRCorner DSRStaircase DSRAverage DSRSpatialJBModel=Gemini-2.5-flash...Model=Gemini-2.5-flash-preview2026.019587.582.572.567.5—SpatialJBModel=ChatGPT-4oModel=ChatGPT-4o2026.0192.5957587.587.5—SpatialJBModel=DeepSeek-V3Model=DeepSeek-V32026.0192.58072.59065—SpatialJBModel=Claude-3-7-sonnetModel=Claude-3-7-sonnet2026.0187.582.577.57572.5—SpatialJBModel=Qwen-longModel=Qwen-long2026.018590958092.5—