ResearchBenchmarksText Safety Filter Bypass on NSFW-200 (ShieldLM 1.0 Target Filter)Follow92.5Bypass RateOptJail92.2493.99595.7597.505May 25, 2025Evaluation ResultsMethodMethodLinksBypass RateOptJailOptimization Filter=GP...Optimization Filter=GPT-4.1, Zero-shot Transfer=true2025.0592.5OptJailOptimization Filter=De...Optimization Filter=DeepSeek-V3, Zero-shot Transfer=true2025.0594OptJailOptimization Filter=Sh...Optimization Filter=ShieldLM, Zero-shot Transfer=false2025.0599