List your company

Prompt-Injection Defense on Adaptive Prompt Injection (train)

32Attack Success Rate (ASR)ACT
29.2847.646684.36May 27, 2026

Evaluation results

MethodLinks
ACTTarget Model=GPT-OSS-20B, Attacker suffixes=best-of-82026.0532
BCTTarget Model=Phi-4-reasoning, Attacker suffixes=best-of-82026.0542.4
ACTTarget Model=Phi-4-reasoning, Attacker suffixes=best-of-82026.0550.2
BCTTarget Model=GPT-OSS-20B, Attacker suffixes=best-of-82026.0552
ACTTarget Model=Qwen3-8B, Attacker suffixes=best-of-82026.0565
ACTTarget Model=Qwen3-1.7B, Attacker suffixes=best-of-82026.0573
BCTTarget Model=Qwen3-1.7B, Attacker suffixes=best-of-82026.0592
BaselineTarget Model=Phi-4-reasoning, Attacker suffixes=best-of-82026.0594
BCTTarget Model=Qwen3-8B, Attacker suffixes=best-of-82026.0595
BaselineTarget Model=Qwen3-8B, Attacker suffixes=best-of-82026.0598
ACTTarget Model=Gemma-4-E4B-it, Attacker suffixes=best-of-82026.0599
BaselineTarget Model=Qwen3-1.7B, Attacker suffixes=best-of-82026.05100
BaselineTarget Model=GPT-OSS-20B, Attacker suffixes=best-of-82026.05100
BaselineTarget Model=Gemma-4-E4B-it, Attacker suffixes=best-of-82026.05100
BCTTarget Model=Gemma-4-E4B-it, Attacker suffixes=best-of-82026.05100