List your company

Safeguarding LLM Agents against Prompt Injection on Banking and Slack (test)

91.9BU (No Attack)Repeated-Prompt
21.59639.84858.176.352May 4, 2026

Evaluation results

MethodLinks
Repeated-PromptLLM backbone=Qwen32026.0591.90.78.90.45.40.389.30.480.40.692.9255.40.8
SpotlightingLLM backbone=Qwen32026.0589.20.45.40.212.50.494.60.391.10.196.40.6600.3
No DefenseLLM backbone=Qwen32026.0586.5010.708.9087.5094.6092.9058.90
MAGELLM backbone=Qwen32026.0586.54.804.103.403.403.902.803.5
Repeated-PromptLLM backbone=Llama-3.32026.0583.87.210.716.321.41717.923.632.113.776.814.331.817
IPIGuardLLM backbone=Qwen32026.0578.43.802.102.101.11.84.51.89.10.73.79
No DefenseLLM backbone=Llama-3.32026.0575.7026.8035.7044.6071.4092.9054.30
SpotlightingLLM backbone=Llama-3.32026.0575.70.230.4025044.60.460.7085.7049.30.1
MELONLLM backbone=Llama-3.32026.0575.73.123.23.219.63.132.14.357.14.678.68.842.14.8
MELONLLM backbone=Qwen32026.0575.73.43.62.53.62.701.737.52.858.96.820.73.3
IPIGuardLLM backbone=Llama-3.32026.0573302.902.804.51.811.83.64.31.15.3
MAGELLM backbone=Llama-3.32026.05734.803.1031.83.502.702.90.43
Repeated-PromptLLM backbone=GPT-4o-mini2026.0564.97.87.14.67.19.4257.116.15.271.415.425.48.3
No DefenseLLM backbone=GPT-4o-mini2026.0562.2014.3021.4042.9069.6098.2049.30
MAGELLM backbone=GPT-4o-mini2026.0559.53.402.8033.43.102.902.70.72.9
SpotlightingLLM backbone=GPT-4o-mini2026.0559.4014.3017.9039.3060.7098.28.746.11.7
MELONLLM backbone=GPT-4o-mini2026.0556.83.903.203.5043.62.802.30.73.2
IPIGuardLLM backbone=GPT-4o-mini2026.0556.84.204.704.605.73.67.70110.76.8
PI-DetectorLLM backbone=Llama-3.32026.0529.70.300.700.717.91.217.9116.11.110.40.9
PI-DetectorLLM backbone=GPT-4o-mini2026.0524.30.300.700.68.91.312.5135.71.511.41
PI-DetectorLLM backbone=Qwen32026.0524.30.200.300.314.30.612.50.53.60.66.10.5