Question Answering on StrategyQA, MMLU, and HotpotQA Combined
0.6334Overall AccuracyJailAgent
Evaluation Results
| Method | Links | |
|---|---|---|
| JailAgentAgent Backbone=Gemini-3.0-pro2026.04 | 0.6334 | |
| JailAgentAgent Backbone=GPT-52026.04 | 0.6308 | |
| JailAgentAgent Backbone=GPT-4o2026.04 | 0.6174 | |
| JailAgentAgent Backbone=Claude-3.5-haiku2026.04 | 0.5846 | |
| AgentPoisonAgent Backbone=Gemini-3.0-pro2026.04 | 0.5835 | |
| AgentPoisonAgent Backbone=GPT-4o2026.04 | 0.5828 | |
| JailAgentAgent Backbone=GPT-3.5-turbo2026.04 | 0.5718 | |
| BadChainAgent Backbone=Gemini-3.0-pro2026.04 | 0.5706 | |
| BadChainAgent Backbone=GPT-4o2026.04 | 0.557 | |
| AgentPoisonAgent Backbone=GPT-52026.04 | 0.5537 | |
| AgentPoisonAgent Backbone=Claude-3.5-haiku2026.04 | 0.5393 | |
| BadChainAgent Backbone=GPT-52026.04 | 0.5281 | |
| AgentPoisonAgent Backbone=GPT-3.5-turbo2026.04 | 0.5256 | |
| BadChainAgent Backbone=Claude-3.5-haiku2026.04 | 0.5103 | |
| JailAgentAgent Backbone=Llama-3.1-70B2026.04 | 0.4955 | |
| BadChainAgent Backbone=GPT-3.5-turbo2026.04 | 0.485 | |
| PAIRAgent Backbone=Gemini-3.0-pro2026.04 | 0.4755 | |
| PAIRAgent Backbone=GPT-52026.04 | 0.4728 | |
| AgentPoisonAgent Backbone=Llama-3.1-70B2026.04 | 0.4651 | |
| PAIRAgent Backbone=GPT-4o2026.04 | 0.4576 | |
| BadChainAgent Backbone=Llama-3.1-70B2026.04 | 0.4425 | |
| PAIRAgent Backbone=Claude-3.5-haiku2026.04 | 0.442 | |
| PAIRAgent Backbone=GPT-3.5-turbo2026.04 | 0.4343 | |
| JailAgentAgent Backbone=ERNIE-3.52026.04 | 0.4066 | |
| PAIRAgent Backbone=Llama-3.1-70B2026.04 | 0.3836 | |
| AgentPoisonAgent Backbone=ERNIE-3.52026.04 | 0.3544 | |
| BadChainAgent Backbone=ERNIE-3.52026.04 | 0.3475 | |
| PAIRAgent Backbone=ERNIE-3.52026.04 | 0.2919 |