List your company

Harmful Refusal on StrongREJECT Adversarial

37.1Attack Success Rate (ASR)SELF-REDTEAM
35.59245.77155.9566.129May 8, 2026

Evaluation results

MethodLinks
SELF-REDTEAMBackbone=Qwen2.5-7B-IT2026.0537.1
Qwen2.5-7B-ITBackbone=Qwen2.5-7B-IT2026.0547
Qwen2.5-3B-ITBackbone=Qwen2.5-3B-IT2026.0549.7
SELF-REDTEAMBackbone=Qwen2.5-3B-IT2026.0555.9
ABSBackbone=Qwen2.5-7B-IT2026.0564.4
Qwen2.5-14B-ITBackbone=Qwen2.5-14B-IT2026.0564.7
ABSBackbone=Qwen2.5-3B-IT2026.0566.9
SELF-REDTEAMBackbone=Qwen2.5-14B-IT2026.0570.1
ABSBackbone=Qwen2.5-14B-IT2026.0574.8