List your company

Adversarial Attack on LLM Agents on GSM8K

97.71ASRInject Agent
-0.341225.114450.5776.0256May 18, 2026

Evaluation results

MethodLinks
Inject AgentDefense Framework=No Defense2026.0597.71—10080
MINJADefense Framework=No Defense2026.0568.86—74.2960.57
MINJADefense Framework=Prompt Filter2026.0565.43—72.8657.71
OEPDefense Framework=No Defense2026.0559.1477.43—54.29
OEPDefense Framework=Prompt Filter2026.0558.5775.71—54.29
MemoryGraftDefense Framework=No Defense2026.0552.57—10047.71
MemoryGraftDefense Framework=Prompt Filter2026.0550.57—10048
AgentPoisonDefense Framework=Prompt Filter2026.0548.86—10041.14
AgentPoisonDefense Framework=No Defense2026.0546.29—10039.71
OEPDefense Framework=LLM Auditor2026.0540.2947.43—36.86
MINJADefense Framework=LLM Auditor2026.0514.86—16.2911.71
AgentPoisonDefense Framework=LLM Auditor2026.055.71—6.294.57
MemoryGraftDefense Framework=LLM Auditor2026.055.43—8.573.14
Inject AgentDefense Framework=LLM Auditor2026.053.71—4.861.43
Inject AgentDefense Framework=Prompt Filter2026.053.43—5.432.86