Categories Companies Research

Loading the SOTA2 catalog…

Explore

Categories Companies Research

Contribute

Join SOTA2 Contact

SOTA2

© 2026 SOTA2

Build in Bellevue, USA

WizWand is now SOTA2 Research

Papers Benchmarks Datasets Tasks

Sabotage detection research benchmarks · SOTA2 Research

Sabotage detection

Benchmarks

Dataset NameSOTA MethodSort

Dataset NameSOTA methodMetricTrendResultsLast Updated

ImpossibleBench

100Average Precision

40

Apr 14, 2026

90.1Average Precision

28

Apr 14, 2026

BigCodeBench Sabotage (reasoning LLM attacker)

Extract-and-Evaluate

0.87log-AUROC

8

Feb 26, 2026

BigCodeBench-Sabotage traditional LLM attacker

Extract-and-Evaluate

0.84log-AUROC

8

Feb 26, 2026

0.87log-AUROC

8

Feb 26, 2026

78log-AUROC

8

Feb 26, 2026

SHADE-Arena Total

70.6Accuracy

3

Jun 8, 2026

SHADE-Arena Charity Allocation

Full-trajectory

91.4Accuracy

3

Jun 8, 2026

SHADE-Arena Performance Reviews

67.3Accuracy

3

Jun 8, 2026

SHADE-Arena Expenditure Accounting

Full-trajectory

87.9Accuracy

3

Jun 8, 2026

SHADE-Arena API Key Calendar

Full-trajectory

88.5Accuracy

3

Jun 8, 2026

SHADE-Arena Jira Tickets

72.5Accuracy

3

Jun 8, 2026

SHADE-Arena Election Tampering

44.4Accuracy

3

Jun 8, 2026

SHADE-Arena Spam Filter

69.5Accuracy

3

Jun 8, 2026

SHADE-Arena American Airlines

83Accuracy

3

Jun 8, 2026

SHADE-Arena Bank Transfer

85.7Accuracy

3

Jun 8, 2026

SHADE-Arena SSN Task

Full-trajectory

96.3Accuracy

3

Jun 8, 2026

DSTS (Distributed Sub-Threshold Sabotage)

0.048Max Power (A)

1

Jun 11, 2026

Popular datasets

SHADE-Arena BigCodeBench TRACE MLE-Sabotage ImpossibleBench

Follow for updates

Join SOTA2 Contact