Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Safety Detection on DoNotAnswer (held-out)
Loading...
97.4
AUROC
Geometry-Lite
86.376
89.238
92.1
94.962
May 18, 2026
AUROC
Updated 13d ago
Evaluation Results
Method
Method
Links
AUROC
Geometry-Lite
Protocol=Leave-one-ben...
2026.05
97.4
TaT-Disp-LSTM
Protocol=Leave-one-ben...
2026.05
96.8
MultiLayer-Linear
Protocol=Leave-one-ben...
2026.05
93.3
Best single-layer probe
Protocol=Leave-one-ben...
2026.05
88
MultiLayer-DIM
Protocol=Leave-one-ben...
2026.05
86.8
Feedback
Search any
task
Search any
task