List your company
ResearchFormerly WizWand

Harmful Refusal on DAN

94.9ASRQwen2.5-3B-IT
7.22829.98952.7575.511Jun 9, 2025Aug 3, 2025Sep 28, 2025Nov 22, 2025Jan 17, 2026Mar 13, 2026May 8, 2026
Updated 3mo ago

Evaluation results

MethodLinks
Qwen2.5-3B-ITBackbone=Qwen2.5-3B-IT2026.0594.9
Llama3.1-8B-IT-ABTraining=Instruction-tuned ablated base2025.0693.7
Qwen2.5-7B-ITBackbone=Qwen2.5-7B-IT2026.0589.7
SELF-REDTEAMBackbone=Qwen2.5-3B-IT2026.0588.5
Qwen2.5-14B-ITBackbone=Qwen2.5-14B-IT2026.0584.5
ABSBackbone=Qwen2.5-7B-IT2026.0575.6
SELF-REDTEAMBackbone=Qwen2.5-7B-IT2026.0572.6
ABSBackbone=Qwen2.5-3B-IT2026.0572.1
ABSBackbone=Qwen2.5-14B-IT2026.0570.6
SELF-REDTEAMBackbone=Qwen2.5-14B-IT2026.0566.1
Defender-OnlyBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1054.2
Llama3.1-8B-ITTraining=Instruction-tuned base2025.0654
Self-PlayBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1053.7
Llama-3.1-8B-ITModel Status=Base Model, Input Condition=adv harm2025.1053.3
Qwen2.5-3B-ITTraining=Instruction-tuned base2025.0651.7
SFTBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1046.8
Defender-Only + SFTBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1045.2
Self-Play + SFTBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1039.6
Llama3.1-8B-IT-AB + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0639.6
Qwen2.5-7B-ITTraining=Instruction-tuned base2025.0639
ELSBase Model=Llama-3.1-8B-IT, Input Condition=adv harm2025.1037.2
Qwen2.5-3B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0633
Llama3.1-8B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0623.9
Qwen2.5-7B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0622.2
Qwen2.5-14B-ITTraining=Instruction-tuned base2025.0621.7
Qwen2.5-14B-IT + SELF-REDTEAMTraining=SELF-REDTEAM fine-tuned2025.0610.6