ResearchBenchmarksJailbreak Detection on AutoDAN (ASR/FPR/Precision Metrics)Follow98ASR (Recall)RLM-JB92.2893.76595.2596.735Feb 18, 2026Evaluation ResultsMethodMethodLinksASR (Recall)FPRPrecisionF1 ScoreRLM-JBLLM backend=GPT-5.2LLM backend=GPT-5.22026.0298298.9998.49RLM-JBLLM backend=GPT-4oLLM backend=GPT-4o2026.02970.599.7498.35RLM-JBLLM backend=DeepSeek-V3.2LLM backend=DeepSeek-V3.22026.0292.5010096.1