List your company

Red-teaming on Mistral-7B

56.7Attack Success Rate (ASR)Adaptive Instruction Composition
23.9432.44540.9549.455Apr 22, 2026

Evaluation results

MethodLinks
Adaptive Instruction CompositionBandit Type=Aggressive, lambda=0.12026.0456.75,673
Adaptive Instruction CompositionBandit Type=Subtle, lambda=0.12026.0436.33,626
WildTeamingAttack Strategy=WildTeaming2026.0425.22,521