ResearchBenchmarksJailbreak Attack Success Rate on AdvBench LLaMA-3.1-70BFollow39.7ASR (SMO, GPT-4o)HMNS4.65213.75122.8531.949Apr 11, 2026Evaluation ResultsMethodMethodLinksASR (SMO, GPT-4o)ASR (SMO, GPT-5)ASR (DPP, GPT-4o)ASR (DPP, GPT-5)ASR (RPO, GPT-4o)ASR (RPO, GPT-5)ASR (PAR, GPT-4o)ASR (PAR, GPT-5)ASR (PAT, GPT-4o)ASR (PAT, GPT-5)ASR (SAF, GPT-4o)ASR (SAF, GPT-5)Average ASR (GPT-4o)Average ASR (GPT-5)HMNS2026.0439.716.252.939.28362.163.736.647.83046.836.655.636.8ArrAttack2026.0433.710.246.933.27756.157.730.641.82440.830.649.630.8Tempest2026.0424184028685050263520332641.728AutoDAN2026.0497201532262016181412918.514.5FITD2026.0463851510106857595.7