List your company

Prompt Injection Attack Detection on Alpaca

100TPRDualSentinel
35.790452.460269.1385.7998Mar 2, 2026

Evaluation results

MethodLinks
DualSentinelBackbone=GPT-4o, Attacks=System, Target Form=OnlyTarget, Sequence Length=Long2026.031004.8
PPLModel=LlaMA2 13B, Attacks=Ignore, Target format=OnlyTarget, Target sequence length=Long2026.03100100
CleanGenModel=LlaMA2 13B, Attacks=Ignore, Target format=OnlyTarget, Target sequence length=Long2026.03100100
DualSentinelModel=LlaMA2 7B, Attacks=Word, Target format=OnlyTarget, Target sequence length=Long2026.0310011.13
DualSentinelModel=LlaMA2 7B, Attacks=Phrase, Target format=OnlyTarget, Target sequence length=Long2026.0310010.54
DualSentinelModel=LlaMA2 7B, Attacks=Sentence, Target format=OnlyTarget, Target sequence length=Long2026.031009.45
DualSentinelModel=LlaMA2 7B, Attacks=Joint, Target format=OnlyTarget, Target sequence length=Long2026.0310010.52
DualSentinelModel=Qwen2.5 7B, Attacks=Word, Target format=OnlyTarget, Target sequence length=Long2026.031007.01
DualSentinelModel=Qwen2.5 7B, Attacks=Phrase, Target format=OnlyTarget, Target sequence length=Long2026.031006.75
DualSentinelModel=Qwen2.5 7B, Attacks=Sentence, Target format=OnlyTarget, Target sequence length=Long2026.031005.99
DualSentinelModel=Qwen2.5 7B, Attacks=Joint, Target format=OnlyTarget, Target sequence length=Long2026.0399.898.08
DualSentinelModel=Qwen2.5 14B, Attacks=Complete, Target format=OnlyTarget, Target sequence length=Long2026.03993
DualSentinelBackbone=GPT-4o, Attacks=Average, Target Form=OnlyTarget, Sequence Length=Long2026.0397.994.8
DualSentinelModel=LlaMA2 13B, Attacks=System, Target format=OnlyTarget, Target sequence length=Long2026.0397.3916
DualSentinelBackbone=GPT-4o-mini, Attacks=System, Target Form=OnlyTarget, Sequence Length=Long2026.0397.123.6
ConfGuardModel=LlaMA2 13B, Attacks=Ignore, Target format=OnlyTarget, Target sequence length=Long2026.039631.7
DualSentinelModel=Qwen2.5 14B, Attacks=Ignore, Target format=OnlyTarget, Target sequence length=Long2026.03963
DualSentinelBackbone=GPT-4o, Attacks=Complete, Target Form=OnlyTarget, Sequence Length=Long2026.0395.984.8
DualSentinelModel=LlaMA2 13B, Attacks=Complete, Target format=OnlyTarget, Target sequence length=Long2026.0394.816
DualSentinelModel=LlaMA2 13B, Attacks=Tree, Target format=OnlyTarget, Target sequence length=Long2026.0394.0416
DualSentinelBackbone=GPT-4o-mini, Attacks=Average, Target Form=OnlyTarget, Sequence Length=Long2026.0392.163.6
DualSentinelModel=Qwen2.5 14B, Attacks=System, Target format=OnlyTarget, Target sequence length=Long2026.03913
DualSentinelBackbone=GPT-4o-mini, Attacks=Complete, Target Form=OnlyTarget, Sequence Length=Long2026.0387.23.6
DualSentinelModel=LlaMA2 13B, Attacks=Ignore, Target format=OnlyTarget, Target sequence length=Long2026.0387.0816
DualSentinelModel=Qwen2.5 14B, Attacks=Tree, Target format=OnlyTarget, Target sequence length=Long2026.03863
ONIONModel=LlaMA2 13B, Attacks=Ignore, Target format=OnlyTarget, Target sequence length=Long2026.0356.667.1
STRIPModel=LlaMA2 13B, Attacks=Ignore, Target format=OnlyTarget, Target sequence length=Long2026.0349.9274
ParagraphModel=LlaMA2 13B, Attacks=Ignore, Target format=OnlyTarget, Target sequence length=Long2026.0338.266