List your company

Indirect Prompt Injection on AlpacaFarm (test)

0Attack Success RateDelim.
-1.92411.06324.0537.037May 25, 2025

Evaluation results

MethodLinks
Delim.Backbone=Llama-3.2-3B, Attack=Naive, Adversarial Training=SFT2025.050
ISEBackbone=Llama-3.2-3B, Attack=Naive, Adversarial Training=SFT2025.050
AIRBackbone=Llama-3.2-3B, Attack=Naive, Adversarial Training=SFT2025.050
Delim.Backbone=Llama-3.2-3B, Attack=Naive, Adversarial Training=DPO2025.050
ISEBackbone=Llama-3.2-3B, Attack=Naive, Adversarial Training=DPO2025.050
AIRBackbone=Llama-3.2-3B, Attack=Naive, Adversarial Training=DPO2025.050
Delim.Backbone=Llama-3.2-3B, Attack=Ignore, Adversarial Training=SFT2025.050
ISEBackbone=Llama-3.2-3B, Attack=Ignore, Adversarial Training=SFT2025.050
AIRBackbone=Llama-3.2-3B, Attack=Ignore, Adversarial Training=SFT2025.050
Delim.Backbone=Llama-3.2-3B, Attack=Ignore, Adversarial Training=DPO2025.050
ISEBackbone=Llama-3.2-3B, Attack=Ignore, Adversarial Training=DPO2025.050
AIRBackbone=Llama-3.2-3B, Attack=Ignore, Adversarial Training=DPO2025.050
AIRBackbone=Llama-3.2-3B, Attack=Completion, Adversarial Training=SFT2025.050
Delim.Backbone=Llama-3.2-3B, Attack=Completion, Adversarial Training=DPO2025.050
ISEBackbone=Llama-3.2-3B, Attack=Completion, Adversarial Training=DPO2025.050
AIRBackbone=Llama-3.2-3B, Attack=Completion, Adversarial Training=DPO2025.050
Delim.Backbone=Llama-3.2-3B, Attack=Escape Sep., Adversarial Training=DPO2025.050
ISEBackbone=Llama-3.2-3B, Attack=Escape Sep., Adversarial Training=DPO2025.050
AIRBackbone=Llama-3.2-3B, Attack=Escape Sep., Adversarial Training=DPO2025.050
Delim.Backbone=Qwen-2.5-7B, Attack=Naive, Adversarial Training=SFT2025.050
AIRBackbone=Qwen-2.5-7B, Attack=Naive, Adversarial Training=SFT2025.050
Delim.Backbone=Qwen-2.5-7B, Attack=Naive, Adversarial Training=DPO2025.050
ISEBackbone=Qwen-2.5-7B, Attack=Naive, Adversarial Training=DPO2025.050
AIRBackbone=Qwen-2.5-7B, Attack=Naive, Adversarial Training=DPO2025.050
Delim.Backbone=Qwen-2.5-7B, Attack=Ignore, Adversarial Training=SFT2025.050
ISEBackbone=Qwen-2.5-7B, Attack=Ignore, Adversarial Training=SFT2025.050
AIRBackbone=Qwen-2.5-7B, Attack=Ignore, Adversarial Training=SFT2025.050
Delim.Backbone=Qwen-2.5-7B, Attack=Ignore, Adversarial Training=DPO2025.050
ISEBackbone=Qwen-2.5-7B, Attack=Ignore, Adversarial Training=DPO2025.050
AIRBackbone=Qwen-2.5-7B, Attack=Ignore, Adversarial Training=DPO2025.050
ISEBackbone=Qwen-2.5-7B, Attack=Completion, Adversarial Training=SFT2025.050
AIRBackbone=Qwen-2.5-7B, Attack=Completion, Adversarial Training=SFT2025.050
Delim.Backbone=Qwen-2.5-7B, Attack=Completion, Adversarial Training=DPO2025.050
ISEBackbone=Qwen-2.5-7B, Attack=Completion, Adversarial Training=DPO2025.050
AIRBackbone=Qwen-2.5-7B, Attack=Completion, Adversarial Training=DPO2025.050
ISEBackbone=Qwen-2.5-7B, Attack=Escape Sep., Adversarial Training=DPO2025.050
AIRBackbone=Qwen-2.5-7B, Attack=Escape Sep., Adversarial Training=DPO2025.050
Delim.Backbone=Llama-3.1-8B, Attack=Naive, Adversarial Training=SFT2025.050
ISEBackbone=Llama-3.1-8B, Attack=Naive, Adversarial Training=SFT2025.050
AIRBackbone=Llama-3.1-8B, Attack=Naive, Adversarial Training=SFT2025.050
Delim.Backbone=Llama-3.1-8B, Attack=Naive, Adversarial Training=DPO2025.050
ISEBackbone=Llama-3.1-8B, Attack=Naive, Adversarial Training=DPO2025.050
AIRBackbone=Llama-3.1-8B, Attack=Naive, Adversarial Training=DPO2025.050
Delim.Backbone=Llama-3.1-8B, Attack=Ignore, Adversarial Training=SFT2025.050
ISEBackbone=Llama-3.1-8B, Attack=Ignore, Adversarial Training=SFT2025.050
AIRBackbone=Llama-3.1-8B, Attack=Ignore, Adversarial Training=SFT2025.050
Delim.Backbone=Llama-3.1-8B, Attack=Ignore, Adversarial Training=DPO2025.050
ISEBackbone=Llama-3.1-8B, Attack=Ignore, Adversarial Training=DPO2025.050
AIRBackbone=Llama-3.1-8B, Attack=Ignore, Adversarial Training=DPO2025.050
Delim.Backbone=Llama-3.1-8B, Attack=Completion, Adversarial Training=SFT2025.050
ISEBackbone=Llama-3.1-8B, Attack=Completion, Adversarial Training=SFT2025.050
AIRBackbone=Llama-3.1-8B, Attack=Completion, Adversarial Training=SFT2025.050
Delim.Backbone=Llama-3.1-8B, Attack=Completion, Adversarial Training=DPO2025.050
ISEBackbone=Llama-3.1-8B, Attack=Completion, Adversarial Training=DPO2025.050
AIRBackbone=Llama-3.1-8B, Attack=Completion, Adversarial Training=DPO2025.050
ISEBackbone=Llama-3.1-8B, Attack=Escape Sep., Adversarial Training=SFT2025.050
AIRBackbone=Llama-3.1-8B, Attack=Escape Sep., Adversarial Training=SFT2025.050
Delim.Backbone=Llama-3.1-8B, Attack=Escape Sep., Adversarial Training=DPO2025.050
ISEBackbone=Llama-3.1-8B, Attack=Escape Sep., Adversarial Training=DPO2025.050
AIRBackbone=Llama-3.1-8B, Attack=Escape Sep., Adversarial Training=DPO2025.050
ISEBackbone=Llama-3.2-3B, Attack=Completion, Adversarial Training=SFT2025.050.5
Delim.Backbone=Llama-3.2-3B, Attack=Escape Sep., Adversarial Training=SFT2025.050.5
ISEBackbone=Llama-3.2-3B, Attack=Escape Sep., Adversarial Training=SFT2025.050.5
AIRBackbone=Llama-3.2-3B, Attack=Escape Sep., Adversarial Training=SFT2025.050.5
ISEBackbone=Qwen-2.5-7B, Attack=Naive, Adversarial Training=SFT2025.050.5
Delim.Backbone=Qwen-2.5-7B, Attack=Escape Sep., Adversarial Training=SFT2025.050.5
ISEBackbone=Qwen-2.5-7B, Attack=Escape Sep., Adversarial Training=SFT2025.050.5
AIRBackbone=Qwen-2.5-7B, Attack=Escape Sep., Adversarial Training=SFT2025.050.5
Delim.Backbone=Qwen-2.5-7B, Attack=Escape Sep., Adversarial Training=DPO2025.050.5
NoneBackbone=Llama-3.1-8B, Attack=Naive, Adversarial Training=None2025.050.5
Delim.Backbone=Llama-3.1-8B, Attack=Escape Sep., Adversarial Training=SFT2025.050.5
NoneBackbone=Llama-3.2-3B, Attack=Naive, Adversarial Training=None2025.051
Delim.Backbone=Llama-3.2-3B, Attack=Completion, Adversarial Training=SFT2025.051
Delim.Backbone=Qwen-2.5-7B, Attack=Completion, Adversarial Training=SFT2025.051
NoneBackbone=Llama-3.2-3B, Attack=Escape Sep., Adversarial Training=None2025.051.4
NoneBackbone=Llama-3.1-8B, Attack=Escape Sep., Adversarial Training=None2025.051.4
AIRBackbone=Qwen-2.5-7B, Attack=GCG, Adversarial Training=DPO2025.051.6
NoneBackbone=Llama-3.2-3B, Attack=Ignore, Adversarial Training=None2025.052.5
NoneBackbone=Llama-3.1-8B, Attack=Ignore, Adversarial Training=None2025.052.5
AIRBackbone=Llama-3.1-8B, Attack=GCG, Adversarial Training=DPO2025.052.8
NoneBackbone=Qwen-2.5-7B, Attack=Ignore, Adversarial Training=None2025.052.9
NoneBackbone=Qwen-2.5-7B, Attack=Escape Sep., Adversarial Training=None2025.052.9
NoneBackbone=Qwen-2.5-7B, Attack=Naive, Adversarial Training=None2025.053.4
NoneBackbone=Llama-3.2-3B, Attack=Completion, Adversarial Training=None2025.053.8
NoneBackbone=Qwen-2.5-7B, Attack=Completion, Adversarial Training=None2025.053.8
NoneBackbone=Llama-3.1-8B, Attack=Completion, Adversarial Training=None2025.053.8
ISEBackbone=Llama-3.1-8B, Attack=GCG, Adversarial Training=DPO2025.054
AIRBackbone=Llama-3.2-3B, Attack=GCG, Adversarial Training=SFT2025.054.1
AIRBackbone=Llama-3.2-3B, Attack=GCG, Adversarial Training=DPO2025.055.2
ISEBackbone=Qwen-2.5-7B, Attack=GCG, Adversarial Training=DPO2025.057.7
AIRBackbone=Llama-3.1-8B, Attack=GCG, Adversarial Training=SFT2025.0511.3
Delim.Backbone=Llama-3.1-8B, Attack=GCG, Adversarial Training=DPO2025.0513
ISEBackbone=Llama-3.1-8B, Attack=GCG, Adversarial Training=SFT2025.0519.9
AIRBackbone=Qwen-2.5-7B, Attack=GCG, Adversarial Training=SFT2025.0522.6
Delim.Backbone=Llama-3.2-3B, Attack=GCG, Adversarial Training=DPO2025.0529.1
Delim.Backbone=Qwen-2.5-7B, Attack=GCG, Adversarial Training=DPO2025.0532
ISEBackbone=Qwen-2.5-7B, Attack=GCG, Adversarial Training=SFT2025.0536.6
Delim.Backbone=Llama-3.2-3B, Attack=GCG, Adversarial Training=SFT2025.0538
ISEBackbone=Llama-3.2-3B, Attack=GCG, Adversarial Training=DPO2025.0546.6
ISEBackbone=Llama-3.2-3B, Attack=GCG, Adversarial Training=SFT2025.0548.1