List your company

Vulnerability Discovery on Meta-prompt-guided adversarial generation corpus GPT-OSS 20B v1.0

10RHOur Framework
0.643.075.57.93Dec 21, 2025

Evaluation results

MethodLinks
Our FrameworkDescription=Automated Red-Teaming Framework using meta-prompt-guided generation2025.121011768547
AdvPrompterDescription=Automated method emphasizing gradient-guided perturbation success2025.1256433223
Template-basedDescription=Static pattern-matching using curated jailbreak templates2025.1245323118
Manual ExpertDescription=Conducted by three AI safety experts following a structured protocol2025.1234212012
Random GenerationDescription=Stochastically samples syntactic templates and n-grams2025.121210105