List your company

Prompt Injection Attacks on OPEN-PROMPT-INJECTIONS

14.2Mean Attack Success Rate (ASR)Llama-3.3-70B-Instruct
11.79628.02344.2560.477Jan 5, 2026

Evaluation results

MethodLinks
Llama-3.3-70B-InstructAttack=ignore2026.0114.215.4-0.130.522
Llama-3.3-70B-InstructAttack=combine2026.0116.517.4-0.1070.607
Llama-3.3-70B-InstructAttack=naive2026.0127.217.9——
Llama-3.1-8B-InstructAttack=fake_completion2026.01389.6-0.0230.943
Llama-3.3-70B-InstructAttack=escape2026.0138.816.90.1161.426
Llama-3.3-70B-InstructAttack=fake_completion2026.0139.814.20.1261.463
Llama-3.1-8B-InstructAttack=naive2026.0140.311.7——
Llama-3.1-8B-InstructAttack=escape2026.0146.711.30.0641.159
Jamba-1.7-MiniAttack=naive2026.014911.7——
Llama-3.1-8B-InstructAttack=ignore2026.0149.86.20.0951.236
Llama-3.1-8B-InstructAttack=combine2026.01517.30.1071.265
Jamba-1.7-MiniAttack=escape2026.0156.58.90.0751.153
Jamba-1.7-MiniAttack=ignore2026.0161.212.20.1221.249
Jamba-1.7-MiniAttack=fake_completion2026.0171.3160.2231.455
Jamba-1.7-MiniAttack=combine2026.0174.316.50.2531.516