Prompt Injection on GSM8k
3.79ASR@3G-Safeguard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| G-SafeguardTopology=Random, Backbone LLM=GPT-4o-mini2025.08 | 3.79 | 0.9911 | |
| G-SafeguardTopology=Chain, Backbone LLM=GPT-4o-mini2025.08 | 4.4 | 0.9822 | |
| G-SafeguardTopology=Tree2025.12 | 6 | 0.9778 | |
| G-SafeguardTopology=Star, Backbone LLM=GPT-4o-mini2025.08 | 6.1 | 0.9822 | |
| BlindGuardTopology=Random, Backbone LLM=GPT-4o-mini2025.08 | 6.44 | 0.7556 | |
| G-SafeguardTopology=Star2025.12 | 6.67 | 0.9911 | |
| DOMINANTTopology=Tree, Backbone LLM=GPT-4o-mini2025.08 | 6.78 | 0.6844 | |
| BlindGuardTopology=Star, Backbone LLM=GPT-4o-mini2025.08 | 6.78 | 0.7022 | |
| G-SafeguardTopology=Tree, Backbone LLM=GPT-4o-mini2025.08 | 7.8 | 0.9911 | |
| DOMINANTTopology=Star, Backbone LLM=GPT-4o-mini2025.08 | 7.8 | 0.6978 | |
| TAMTopology=Tree, Backbone LLM=GPT-4o-mini2025.08 | 8.14 | 0.5467 | |
| TAMTopology=Star, Backbone LLM=GPT-4o-mini2025.08 | 8.14 | 0.6844 | |
| DOMINANTTopology=Chain, Backbone LLM=GPT-4o-mini2025.08 | 8.47 | 0.6756 | |
| BlindGuardTopology=Chain, Backbone LLM=GPT-4o-mini2025.08 | 8.47 | 0.6933 | |
| PREMTopology=Tree, Backbone LLM=GPT-4o-mini2025.08 | 8.47 | 0.5333 | |
| BlindGuardTopology=Tree, Backbone LLM=GPT-4o-mini2025.08 | 8.47 | 0.5955 | |
| XG-GuardTopology=Chain2025.12 | 8.67 | 0.9778 | |
| PREMTopology=Chain, Backbone LLM=GPT-4o-mini2025.08 | 8.79 | 0.6222 | |
| TAMTopology=Chain, Backbone LLM=GPT-4o-mini2025.08 | 8.84 | 0.5156 | |
| PREMTopology=Tree2025.12 | 8.89 | 0.6296 | |
| G-SafeguardTopology=Chain2025.12 | 9.33 | 0.9911 | |
| XG-GuardTopology=Tree2025.12 | 9.67 | 0.9733 | |
| TAMTopology=Star2025.12 | 9.67 | 0.7111 | |
| G-SafeguardTopology=Random2025.12 | 9.67 | 0.9911 | |
| DOMINANTTopology=Star2025.12 | 9.83 | 0.6701 | |
| No DefenseTopology=Chain, Backbone LLM=GPT-4o-mini2025.08 | 9.83 | — | |
| DOMINANTTopology=Tree2025.12 | 10.18 | 0.6897 | |
| No DefenseTopology=Tree, Backbone LLM=GPT-4o-mini2025.08 | 10.2 | — | |
| PREMTopology=Chain2025.12 | 10.29 | 0.5962 | |
| BlindGuardTopology=Star2025.12 | 10.33 | 0.7067 | |
| XG-GuardTopology=Random2025.12 | 10.33 | 0.9867 | |
| DOMINANTTopology=Random2025.12 | 10.51 | 0.6881 | |
| PREMTopology=Star, Backbone LLM=GPT-4o-mini2025.08 | 10.51 | 0.5956 | |
| DOMINANTTopology=Random, Backbone LLM=GPT-4o-mini2025.08 | 10.51 | 0.6978 | |
| PREMTopology=Random, Backbone LLM=GPT-4o-mini2025.08 | 10.51 | 0.6978 | |
| BlindGuardTopology=Chain2025.12 | 10.67 | 0.6533 | |
| DOMINANTTopology=Chain2025.12 | 10.91 | 0.6655 | |
| TAMTopology=Chain2025.12 | 11.67 | 0.5289 | |
| TAMTopology=Tree2025.12 | 11.67 | 0.5467 | |
| XG-GuardTopology=Star2025.12 | 11.67 | 0.9733 | |
| No DefenseTopology=Star, Backbone LLM=GPT-4o-mini2025.08 | 12.89 | — | |
| PREMTopology=Star2025.12 | 13.08 | 0.641 | |
| No DefenseTopology=Chain2025.12 | 13.67 | — | |
| No DefenseTopology=Tree2025.12 | 13.67 | — | |
| BlindGuardTopology=Tree2025.12 | 14 | 0.5599 | |
| BlindGuardTopology=Random2025.12 | 14.33 | 0.7422 | |
| No DefenseTopology=Random, Backbone LLM=GPT-4o-mini2025.08 | 14.48 | — | |
| TAMTopology=Random, Backbone LLM=GPT-4o-mini2025.08 | 14.91 | 0.4622 | |
| No DefenseTopology=Star2025.12 | 15 | — | |
| TAMTopology=Random2025.12 | 15.67 | 0.4844 | |
| PREMTopology=Random2025.12 | 16.14 | 0.6257 | |
| No DefenseTopology=Random2025.12 | 19.33 | — |