List your company

Refusal on Llama-3-8B n≈200

0ASRFine-tune
-2.3213.342944.66Apr 27, 2026

Evaluation results

MethodLinks
Fine-tuneAttack Type=BadNet2026.0409—
CleanGenAttack Type=BadNet2026.0400.5—
CleanGenAttack Type=VPI2026.0401—
LCFAttack Type=VPI2026.040—10.5
CROWAttack Type=Sleeper2026.04028.4—
CleanGenAttack Type=StyleBkd2026.0400.5—
PruningAttack Type=CTBA2026.04085—
CleanGenAttack Type=MTBA2026.0401—
LCFAttack Type=MTBA2026.040—8.5
PruningAttack Type=BadNet2026.040.578.5—
PruningAttack Type=Sleeper2026.040.582.5—
CleanGenAttack Type=Sleeper2026.040.51.5—
CleanGenAttack Type=SynBkd2026.040.51.5—
LCFAttack Type=SynBkd2026.040.5—11
Fine-tuneAttack Type=CTBA2026.040.512—
CleanGenAttack Type=CTBA2026.040.51.5—
Fine-tuneAttack Type=Sleeper2026.04112.5—
PruningAttack Type=SynBkd2026.04169.5—
LCFAttack Type=StyleBkd2026.041.1—9.5
PruningAttack Type=StyleBkd2026.041.668—
CROWAttack Type=BadNet2026.04226.4—
CROWAttack Type=SynBkd2026.043.514.9—
PruningAttack Type=VPI2026.044.582—
DecodingAttack Type=SynBkd2026.0453—
LCFAttack Type=Sleeper2026.045.5—12.5
PruningAttack Type=MTBA2026.045.581.5—
DecodingAttack Type=Sleeper2026.0461—
Fine-tuneAttack Type=SynBkd2026.046.57—
CROWAttack Type=StyleBkd2026.046.523.9—
DecodingAttack Type=BadNet2026.0490.5—
CROWAttack Type=CTBA2026.0411.524.4—
LCFAttack Type=BadNet2026.0412—13
Fine-tuneAttack Type=MTBA2026.0413.58.5—
CROWAttack Type=MTBA2026.0413.521.9—
Fine-tuneAttack Type=StyleBkd2026.04145.5—
Fine-tuneAttack Type=VPI2026.0419.513.5—
DecodingAttack Type=CTBA2026.04281—
LCFAttack Type=CTBA2026.0430—10
DecodingAttack Type=VPI2026.0430.53.5—
CROWAttack Type=VPI2026.043326.4—
DecodingAttack Type=StyleBkd2026.0445.21—
DecodingAttack Type=MTBA2026.04583.5—