List your company

Safety Evaluation on 8-first-plot

0Misalignment Ratere-SFT
-2.46814.19130.8547.509Jun 8, 2026

Evaluation results

MethodLinks
re-SFTMitigation Strategy=Realignment Methods, Training Free=false2026.060——————————————
Gating InversionMitigation Strategy=Realignment Methods, Training Free=true2026.060——————————————
Activation SteerMitigation Strategy=Realignment Methods, Training Free=true2026.067——————————————
Misaligned LoRAMitigation Strategy=w/o Mitigation2026.0661.7——————————————
Alignment Gating (Gating Inversion)Backbone=Qwen3-8B, Protocol=Gating Inversion, Format=Free-form generation2026.06——0000000000000
Alignment Gating (Gating Inversion)Backbone=Qwen3-14B, Protocol=Gating Inversion, Format=Free-form generation2026.06——0000000000000