List your company

Prompt Injection Attack Defense on BIPIA

0ASRPrompt Guard
-0.05320.30590.6651.0241Apr 1, 2026Apr 10, 2026Apr 20, 2026Apr 30, 2026May 9, 2026May 19, 2026May 29, 2026

Evaluation results

MethodLinks
Prompt Guard2026.0400.339.221,077.71——
SelfDefendMode=Direct2026.04019.330.413,113.04——
SelfGrader2026.0405.330.3328.91——
SCOUTJudge=GPT-4o2026.050.026—0.309—97.196.8
Always-DLLMJudge=GPT-4o2026.050.044—1.654—89.483.2
SCOUTJudge=GPT-5.12026.050.058—0.433—96.198
Always-DLLMJudge=GPT-5.12026.050.236—1.586—87.398.2
GradientCuff2026.040.3339.3310.49827.33——
SelfDefendMode=Intent2026.040.3367.661.2613,116.22——
WildGuardMode=Pre2026.040.6681.661.5827,861.87——
WildGuardMode=Post2026.040.66891.4927,882.3——
GuardReasonerMode=Pre2026.04189.338.5815,414.55——
GuardReasonerMode=Post2026.04197.668.2215,414.6——
LLama-3-8B-InstructDefense=No Defense2026.041.33—————
Perplexity Filter2026.041.331000.0913,112.46——
Llama GuardMode=Pre2026.041.331000.2213,571.94——
Llama GuardMode=Post2026.041.331000.2213,598.72——