List your company

Safety Jailbreak Evaluation on Forbidden Questions

53.6ASRBaseline (No Attack)
52.05262.50172.9583.399Mar 10, 2026

Evaluation results

MethodLinks
Baseline (No Attack)Model=Llama-2-7B-Chat, Setting=Baseline (No Attack)2026.0353.6
Baseline (No Attack)Model=Llama-3-8B-Instruct, Setting=Baseline (No Attack)2026.0369.2
AmnesiaModel=Llama-3-8B-Instruct, Setting=Amnesia (Layers 13,14,15)2026.0392.3