List your company

ResearchFormerly WizWand

Harmful Request Compliance

Benchmarks

DatasetSOTA methodMetricTrendResultsLast updated

Harmful Requests RepE Attack

Llama-3-8B-Instruct

91.2Compliance Rate

6

Feb 26, 2026

Harmful Requests Input Embed

Mistral-7B-Instruct-v2

92.1Compliance Rate

6

Feb 26, 2026

Harmful Requests Prefilling

Mistral-7B-Instruct-v2

95Compliance Rate

6

Feb 26, 2026

Harmful Requests Multilingual

Mistral-7B-Instruct-v2

34.1Compliance Rate

6

Feb 26, 2026

Harmful Requests GCG

Mistral-7B-Instruct-v2

0.887Compliance Rate

6

Feb 26, 2026

Harmful Requests PAIR

Mistral-7B-Instruct-v2

69.5Compliance Rate

6

Feb 26, 2026

Harmful Requests TAP-T

Mistral-7B-Instruct-v2

85.8Compliance Rate

6

Feb 26, 2026

Harmful Requests AutoDAN

Mistral-7B-Instruct-v2

93.4Compliance Rate

6

Feb 26, 2026

Harmful Requests Manual Attack

Mistral-7B-Instruct-v2

77.4Compliance Rate

6

Feb 26, 2026

Harmful Requests No Attack

Mistral-7B-Instruct-v2

57.8Compliance Rate

6

Feb 26, 2026