List your company

Jailbreak evaluation on Synthetic dataset (held-out)

100Good Bot RateAIM
-2.9623.7750.577.23Jul 5, 2023

Evaluation results

MethodLinks
AIMModel=Claude v1.32023.07100—0
AIMModel=GPT-42023.0713—1
combination_3Model=Claude v1.32023.0712—2
Adaptive attackModel=GPT-42023.07496—
combination_2Model=GPT-42023.073—10
combination_2Model=Claude v1.32023.073—8
combination_3Model=GPT-42023.072—5
Adaptive attackModel=Claude v1.32023.07199—