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

Instruction Hierarchy

Benchmarks

Task NameDataset NameSOTA ResultTrend
Instruction HierarchyInstruction Hierarchy Tutor Jailbreaks
Pass Rate95
4
Phrase ProtectionInstruction Hierarchy (test)
User Message Protection Accuracy97.5
4
System Prompt ExtractionInstruction Hierarchy (test)
Attack Success Rate (Realistic User)99.7
4
Instruction Following SafetyInstruction Hierarchy Phrase and Password Protection
Phrase Protection Adherence (User)91
2
Showing 4 of 4 rows