List your company

Text Safety Filter Bypass on NSFW-200 Target Filter: GPT-4.1 1.0 (cross-filter transfer)

97.5Bypass RateOptJail
94.995.57596.2596.925May 25, 2025

Evaluation results

MethodLinks
OptJailOptimization Filter=GPT-4.1, Zero-shot Transfer=false2025.0597.5
OptJailOptimization Filter=ShieldLM, Zero-shot Transfer=true2025.0597
OptJailOptimization Filter=DeepSeek-V3, Zero-shot Transfer=true2025.0595