List your company

Text Safety Filter Bypass on NSFW-200 (ShieldLM 1.0 Target Filter)

92.5Bypass RateOptJail
92.2493.99595.7597.505May 25, 2025

Evaluation results

MethodLinks
OptJailOptimization Filter=GPT-4.1, Zero-shot Transfer=true2025.0592.5
OptJailOptimization Filter=DeepSeek-V3, Zero-shot Transfer=true2025.0594
OptJailOptimization Filter=ShieldLM, Zero-shot Transfer=false2025.0599