List your company

Indirect Prompt Injection on HotpotQA

100ASRObliInjection-CE
-3.79223.15450.177.046Dec 10, 2025

Evaluation results

MethodLinks
ObliInjection-CETarget LLM=Mistral-7B2025.12100—
ObliInjection-CETarget LLM=Qwen-2.5-7B2025.12100—
ObliInjectionTarget LLM=Llama-3-8B2025.12100—
ObliInjectionTarget LLM=Mistral-7B2025.12100—
ObliInjectionTarget LLM=Falcon3-7B2025.12100—
ObliInjectionTarget LLM=Qwen3-4B2025.12100—
ObliInjection-CETarget LLM=Llama4-17B2025.1299.8—
ObliInjectionTarget LLM=Llama4-17B2025.1299.6—
ObliInjectionTarget LLM=Llama-3.1-8B2025.1299.2—
ObliInjectionTarget LLM=Qwen-2.5-7B2025.1298.6—
JudgeDeceiverTarget LLM=Qwen-2.5-7B2025.1293.4—
ObliInjection-CETarget LLM=Llama-3.1-8B2025.1293—
ObliInjection-GCGTarget LLM=Mistral-7B2025.1292.6—
ObliInjection-CETarget LLM=Falcon3-7B2025.1291.6—
ObliInjection-GCGTarget LLM=Falcon3-7B2025.1288.6—
JudgeDeceiverTarget LLM=Mistral-7B2025.1288—
ObliInjection-GCGTarget LLM=Llama-3-8B2025.1285.4—
ObliInjection-CETarget LLM=Qwen3-4B2025.1284.6—
Combined AttackTarget LLM=Qwen-2.5-7B2025.1284.4—
ObliInjection-GCGTarget LLM=Qwen-2.5-7B2025.1270—
ObliInjection-CETarget LLM=Llama-3-8B2025.1267.8—
JudgeDeceiverTarget LLM=Falcon3-7B2025.1258.8—
JudgeDeceiverTarget LLM=Llama-3-8B2025.1255.6—
Combined AttackTarget LLM=Llama-3.1-8B2025.1239.6—
Neural ExecTarget LLM=Llama-3-8B2025.1232.8—
JudgeDeceiverTarget LLM=Llama-3.1-8B2025.1232.8—
ObliInjection-GCGTarget LLM=Llama-3.1-8B2025.1231.2—
Combined AttackTarget LLM=Qwen3-4B2025.1228.4—
Neural ExecTarget LLM=Qwen3-4B2025.1227.8—
Neural ExecTarget LLM=Mistral-7B2025.1225.5—
Neural ExecTarget LLM=Llama-3.1-8B2025.1225.4—
Neural ExecTarget LLM=Falcon3-7B2025.1220.2—
Combined AttackTarget LLM=Mistral-7B2025.1220—
Combined AttackTarget LLM=Llama-3-8B2025.1214.2—
ObliInjection-GCGTarget LLM=Qwen3-4B2025.1214—
JudgeDeceiverTarget LLM=Qwen3-4B2025.1210.6—
Neural ExecTarget LLM=Llama4-17B2025.128.8—
ObliInjection-GCGTarget LLM=Llama4-17B2025.128.2—
Neural ExecTarget LLM=Qwen-2.5-7B2025.122.8—
Combined AttackTarget LLM=Llama4-17B2025.121.6—
Combined AttackTarget LLM=Falcon3-7B2025.120.4—
JudgeDeceiverTarget LLM=Llama4-17B2025.120.2—
Combined AttackTarget LLM=Average2025.12—26.9
JudgeDeceiverTarget LLM=Average2025.12—48.5
Neural ExecTarget LLM=Average2025.12—20.5
ObliInjectionTarget LLM=Average2025.12—99.6
ObliInjection-CETarget LLM=Average2025.12—91
ObliInjection-GCGTarget LLM=Average2025.12—55.7