Our new X account is live! Follow @wizwand_team for updates
Home
/
Benchmarks
Sabotage detection on MLE-Sabotage
Loading...
0.87
log-AUROC
Action-only
0.4748
0.5774
0.68
0.7826
Jan 28, 2026
log-AUROC
Best Monitor Type
Updated 4d ago
Evaluation Results
Method
Method
Links
log-AUROC
Best Monitor Type
Action-only
Monitor Model=GPT-4.1
2026.01
0.87
-
Extract-and-Evaluate
Monitor Model=Claude-3...
2026.01
0.81
-
Extract-and-Evaluate
Monitor Model=GPT-4.1
2026.01
0.8
-
CoT-only
Monitor Model=Claude-3...
2026.01
0.72
-
CoT+action
Monitor Model=GPT-4.1
2026.01
0.58
-
CoT-only
Monitor Model=GPT-4.1
2026.01
0.57
-
Action-only
Monitor Model=Claude-3...
2026.01
0.54
-
CoT+action
Monitor Model=Claude-3...
2026.01
0.49
-
Feedback
Search any
task
Search any
task