List your company

Adversarial Attack on AdvBench

100Attack Success Rate (ASR)Standard
-3.771223.169450.1177.0506Feb 13, 2025May 5, 2025Jul 26, 2025Oct 15, 2025Jan 5, 2026Mar 27, 2026Jun 17, 2026

Evaluation results

MethodLinks
StandardTarget Model=Llama-3.1-8B, Initialization=Standard2025.021002122.023.16——
BaselineTarget Model=Llama-3.1-8B, Initialization=Baseline2025.0210033.961.58——
1-CRITarget Model=Llama-3.1-8B, Initialization=1-CRI (ours)2025.021007.57.66.59——
5-CRITarget Model=Llama-3.1-8B, Initialization=5-CRI (ours)2025.0210022.520.69——
25-CRITarget Model=Llama-3.1-8B, Initialization=25-CRI (ours)2025.0210077.844.06——
BaselineTarget Model=Mistral-7B, Initialization=Baseline2025.0210048.282.2——
1-CRITarget Model=Mistral-7B, Initialization=1-CRI (ours)2025.0210055.54.02——
5-CRITarget Model=Mistral-7B, Initialization=5-CRI (ours)2025.0210022.30.41——
25-CRITarget Model=Mistral-7B, Initialization=25-CRI (ours)2025.0210077.022.89——
StandardTarget Model=Mistral-7B, Initialization=Standard2025.02921427.262.71——
BaselineJudge=Heuristic2026.0665.18—————
SafeLMJudge=Heuristic2026.0655.56—————
BaselineJudge=LLM2026.0643.99—————
SafeLMJudge=LLM2026.0623.55—————
SRPJudge=Heuristic2026.061.98—————
SRPJudge=LLM2026.060.22—————
Mistral-7B-Instruct-v0.3Attack=PAIR2026.06————15.7713.85
Mistral-7B-Instruct-v0.3Attack=Prefilling2026.06————52.8149.06
Mistral-7B-Instruct-v0.3Attack=I-GCG2026.06————20.9615.19
Mistral-7B-Instruct-v0.3 + Trajectory AlignmentAttack=PAIR2026.06————0.970.19
Mistral-7B-Instruct-v0.3 + Trajectory AlignmentAttack=Prefilling2026.06————77.692.12
Mistral-7B-Instruct-v0.3 + Trajectory AlignmentAttack=I-GCG2026.06————0.380.38