Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Prompt Injection Defense on PI (CSQA) random topology

50ASR @1

No Defense

10.79220.97131.1541.329Jan 21, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2026.01
5070.779.351.728.316.7
2026.01
47.770.778552516.7
2026.01
47.370.77758.326.718.3
2026.01
46.565.376.351.73521.7
2026.01
44.665.475.8553521.7
2026.01
43.75155.356.75046.7
2026.01
37.548.955.663.348.340
2026.01
3745.34961.753.351.7
2026.01
3650.259.365.350.341.7
2026.01
31.730.731.766.77068.3
2026.01
27.227.126.969.273.173.1
2026.01
25.34145.963.35551.7
2026.01
24.323.323.37576.776.7
2026.01
21.156.372.858.328.318.3
2026.01
2124.326.383.376.778.3
2026.01
12.312.713.49086.786.7