List your company

Jailbreaking on EHRAgent eICU

57.93Success Rate (SR)Non-attack
33.54239.873546.20552.5365Apr 7, 2026

Evaluation results

MethodLinks
Non-attackAgent Backbone=Claude-3.5-haiku2026.0457.93———98.1
JailAgentAgent Backbone=Claude-3.5-haiku2026.0457.2495.6991.219097.76
BadChainAgent Backbone=Gemini-3.0-pro2026.0451.7280.3478.6276.5579.66
AgentPoisonAgent Backbone=Claude-3.5-haiku2026.0451.3887.7681.0380.3496.38
Non-attackAgent Backbone=Gemini-3.0-pro2026.0450.86———80.34
JailAgentAgent Backbone=Gemini-3.0-pro2026.0450.5291.7286.5585.8679.48
Non-attackAgent Backbone=GPT-4o2026.0449.66———69.31
BadChainAgent Backbone=Claude-3.5-haiku2026.0448.186.2178.177.4195.52
JailAgentAgent Backbone=GPT-4o2026.0447.9393.4591.991.3869.31
AgentPoisonAgent Backbone=Gemini-3.0-pro2026.0447.7680.8677.4175.6976.9
PAIRAgent Backbone=Claude-3.5-haiku2026.0447.2473.9770.5267.0793.79
JailAgentAgent Backbone=Llama-3.1-70B2026.0446.5582.5978.6276.7281.03
Non-attackAgent Backbone=GPT-52026.0446.38———63.28
Non-attackAgent Backbone=Llama-3.1-70B2026.0446.38———80.17
AgentPoisonAgent Backbone=GPT-4o2026.0446.2187.7685.5282.5962.07
AgentPoisonAgent Backbone=GPT-52026.0445.6986.2179.1477.4163.28
BadChainAgent Backbone=GPT-4o2026.0445.3486.7283.9782.0761.38
JailAgentAgent Backbone=GPT-52026.0445.3491.2189.3189.4862.59
Non-attackAgent Backbone=GPT-3.5-turbo2026.0444.48———61.72
JailAgentAgent Backbone=GPT-3.5-turbo2026.0443.7990.3484.3180.8661.21
BadChainAgent Backbone=Llama-3.1-70B2026.0443.2880.5275.6975.5279.31
AgentPoisonAgent Backbone=Llama-3.1-70B2026.0442.9375.8671.2170.6978.79
PAIRAgent Backbone=GPT-4o2026.0442.7674.6672.7670.8655.17
BadChainAgent Backbone=GPT-52026.0442.2485.8676.5574.6657.93
PAIRAgent Backbone=Gemini-3.0-pro2026.0442.2470.8668.976567.59
AgentPoisonAgent Backbone=GPT-3.5-turbo2026.0442.0784.1478.9778.4556.21
BadChainAgent Backbone=GPT-3.5-turbo2026.0440.8683.6277.5977.9353.28
PAIRAgent Backbone=Llama-3.1-70B2026.0439.4866.7261.958.4577.93
PAIRAgent Backbone=GPT-52026.0436.7275.3469.1469.1452.76
PAIRAgent Backbone=GPT-3.5-turbo2026.0434.4867.936562.4149.48