List your company

Jailbreaking Safety Evaluation on Fortress

87.7Safety ScoreSelf-ReSET
46.41257.13167.8578.569Oct 1, 2025Nov 15, 2025Dec 30, 2025Feb 14, 2026Mar 31, 2026May 15, 2026Jun 30, 2026

Evaluation results

MethodLinks
Self-ReSETBase Model=DeepSeek-R1-Distill-Llama-8B2026.0587.7
RECAPBackbone=DSLlama-8B2025.1086.84
Self-ReSETBase Model=DeepSeek-R1-Distill-Qwen-7B2026.0582.4
RECAPBase Model=DeepSeek-R1-Distill-Llama-8B2026.0581.5
SEARModel Size=1.5B, Pre-fill attack protocol=true2026.0681.5
RECAPBackbone=DSQwen-14B2025.1080.17
STARBackbone=DSQwen-14B2025.1079.05
STARBackbone=DSLlama-8B2025.1078.49
Self-ReSETBase Model=Qwen3-8B2026.0577.2
SFTBackbone=DSQwen-14B2025.1074.54
SFTBackbone=DSLlama-8B2025.1074.25
DAPOBase Model=DeepSeek-R1-Distill-Llama-8B2026.0574.2
STAR-1Base Model=Qwen3-8B2026.0572.9
STAR-1Base Model=DeepSeek-R1-Distill-Llama-8B2026.0572.6
RECAPBase Model=DeepSeek-R1-Distill-Qwen-7B2026.0569.5
STAR-1Base Model=DeepSeek-R1-Distill-Qwen-7B2026.0569
DAPOBackbone=DSLlama-8B2025.1068.86
DAPOBackbone=DSQwen-14B2025.1067.85
DAPOBase Model=DeepSeek-R1-Distill-Qwen-7B2026.0566.4
QWEN3 32BModel Size=32B, Pre-fill attack protocol=true2026.0665.9
SafeChainBackbone=DSLlama-8B2025.1065.56
QWEN3 14BModel Size=14B, Pre-fill attack protocol=true2026.0665.5
QWEN3 8BModel Size=8B, Pre-fill attack protocol=true2026.0663.8
SafeChainBackbone=DSQwen-14B2025.1063.15
DAPOBase Model=Qwen3-8B2026.0562.7
RECAPBase Model=Qwen3-8B2026.0562.5
OriginalBackbone=DSLlama-8B2025.1060.83
OriginalBackbone=DSQwen-14B2025.1057.91
SafechainBase Model=DeepSeek-R1-Distill-Llama-8B2026.0557.3
SafechainBase Model=Qwen3-8B2026.0556
SafechainBase Model=DeepSeek-R1-Distill-Qwen-7B2026.0553.5
BaseBase Model=DeepSeek-R1-Distill-Qwen-7B2026.0549.6
BaseBase Model=Qwen3-8B2026.0548.9
BaseBase Model=DeepSeek-R1-Distill-Llama-8B2026.0548