Video Question Answering on VideoAgent Aggregate: EgoSchema and NExT-QA
68.606Overall AccuracyJailAgent
Evaluation Results
| Method | Links | |
|---|---|---|
| JailAgentAgent Backbone=Gemini-3.0-pro2026.04 | 68.606 | |
| JailAgentAgent Backbone=GPT-52026.04 | 66.346 | |
| JailAgentAgent Backbone=GPT-4o2026.04 | 63.849 | |
| JailAgentAgent Backbone=Claude-3.5-haiku2026.04 | 63.815 | |
| BadChainAgent Backbone=Gemini-3.0-pro2026.04 | 63.193 | |
| AgentPoisonAgent Backbone=Gemini-3.0-pro2026.04 | 61.886 | |
| JailAgentAgent Backbone=GPT-3.5-turbo2026.04 | 61.366 | |
| JailAgentAgent Backbone=Llama-3.1-70B2026.04 | 61.057 | |
| AgentPoisonAgent Backbone=GPT-4o2026.04 | 61.034 | |
| AgentPoisonAgent Backbone=Claude-3.5-haiku2026.04 | 60.661 | |
| AgentPoisonAgent Backbone=GPT-52026.04 | 59.351 | |
| BadChainAgent Backbone=GPT-52026.04 | 59.159 | |
| BadChainAgent Backbone=GPT-3.5-turbo2026.04 | 59.064 | |
| BadChainAgent Backbone=Llama-3.1-70B2026.04 | 57.666 | |
| BadChainAgent Backbone=GPT-4o2026.04 | 57.568 | |
| AgentPoisonAgent Backbone=Llama-3.1-70B2026.04 | 57.55 | |
| AgentPoisonAgent Backbone=GPT-3.5-turbo2026.04 | 57.454 | |
| PAIRAgent Backbone=Gemini-3.0-pro2026.04 | 55.839 | |
| BadChainAgent Backbone=Claude-3.5-haiku2026.04 | 54.834 | |
| PAIRAgent Backbone=GPT-4o2026.04 | 52.648 | |
| PAIRAgent Backbone=GPT-52026.04 | 52.573 | |
| PAIRAgent Backbone=Claude-3.5-haiku2026.04 | 50.456 | |
| PAIRAgent Backbone=Llama-3.1-70B2026.04 | 50.082 | |
| PAIRAgent Backbone=GPT-3.5-turbo2026.04 | 49.186 |