List your company

Semantic Alignment on Safety Concepts Multi-Concept

-0.0396Delta OriginalUndefended Model
-0.04158-0.04059-0.0396-0.03861Feb 22, 2026

Evaluation results

MethodLinks
Undefended Model2026.02-0.0396———
ReVision2026.02—0.01980.05930.0458