Embodied Agent Planning (Adversarial Safety Evaluation) on SafeAgentBench Unsafe Tasks - Jailbreak
54.85SRLota-Bench
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Lota-BenchBackbone=GPT-4, Safety Prompting (Safety-LLM)=false, Execution Module=Identical2026.05 | 54.85 | 0 | 19.87 | |
| LaMMA-PBackbone=GPT-4, Safety Prompting (Safety-LLM)=false, Execution Module=Identical2026.05 | 42.14 | 0 | 28.24 | |
| MLDTBackbone=GPT-4, Safety Prompting (Safety-LLM)=false, Execution Module=Identical2026.05 | 40.13 | 0 | 26.33 | |
| ProgPromptBackbone=GPT-4, Safety Prompting (Safety-LLM)=false, Execution Module=Identical2026.05 | 38.13 | 0 | 27.9 | |
| ReActBackbone=GPT-4, Safety Prompting (Safety-LLM)=false, Execution Module=Identical2026.05 | 35.12 | 2.01 | 25.52 | |
| Lota-Bench (+ Safety LLM)Backbone=GPT-4, Safety Prompting (Safety-LLM)=true, Execution Module=Identical2026.05 | 25.08 | 51.17 | 16.21 | |
| ReAct (+ Safety LLM)Backbone=GPT-4, Safety Prompting (Safety-LLM)=true, Execution Module=Identical2026.05 | 18.06 | 49.16 | 21.05 | |
| LaMMA-P (+ Safety LLM)Backbone=GPT-4, Safety Prompting (Safety-LLM)=true, Execution Module=Identical2026.05 | 16.05 | 52.17 | 23.07 | |
| MLDT (+ Safety LLM)Backbone=GPT-4, Safety Prompting (Safety-LLM)=true, Execution Module=Identical2026.05 | 15.05 | 42.81 | 23.4 | |
| ProgPrompt (+ Safety LLM)Backbone=GPT-4, Safety Prompting (Safety-LLM)=true, Execution Module=Identical2026.05 | 12.04 | 44.82 | 22.64 | |
| NEXUS (0-shot)Backbone=GPT-4, Knowledge Evolution Status=0-shot2026.05 | 1.34 | 62.88 | 57.19 | |
| NEXUS (Evolved)Backbone=GPT-4, Knowledge Evolution Status=Evolved2026.05 | 0.67 | 72.24 | 34.32 |