EvoJail uses evolutionary algorithms with field-aware instruction fusion and multi-level mutations to generate adaptable, diverse jailbreak prompts for LLMs, claiming over 93% attack success rate and 5.6% diversity gains over prior methods.
{harm_instruction}
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.NE 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
EvoJail: Evolutionary Diverse Jailbreak Prompt Generation for Large Language Models
EvoJail uses evolutionary algorithms with field-aware instruction fusion and multi-level mutations to generate adaptable, diverse jailbreak prompts for LLMs, claiming over 93% attack success rate and 5.6% diversity gains over prior methods.