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

ClearHarm

Benchmarks

Task NameDataset NameSOTA ResultTrend
Jailbreak AttackClearHarm
ASR61
35
Jailbreak RefusalClearHarm jailbreak refusal
Refusal Rate (ClearHarm)87
6
Answer SafetyClearHarm
Answer Safety Score90.1
4
Prefill Attack DefenseClearHarm prefill (held-out)
PAR52
4
Jailbreak DefenseClearHarm Jailbreak (held-out)
Attack Success Rate (ASR)47
4
Showing 5 of 5 rows