Jailbreak Attack Evaluation on TRIDENT CORE
7HPRTRIDENT-CORE
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TRIDENT-COREModel=Llama-3.1-8B-chat, Attack Variant=Standard2025.05 | 7 | 1.72 | 1 | |
| TRIDENT-COREModel=GPT-4O-MINI, Attack Variant=+ RENELLM2025.05 | 8 | 1.99 | 1 | |
| TRIDENT-COREModel=GPT-4O-MINI, Attack Variant=Standard2025.05 | 8 | 1.99 | 1 | |
| TRIDENT-COREModel=GPT-4O-MINI, Attack Variant=+ Code Injection2025.05 | 14 | 2.28 | 12 | |
| TRIDENT-COREModel=GPT-4O-MINI, Attack Variant=Instructions2025.05 | 23 | 2.32 | 12 | |
| TRIDENT-COREModel=GPT-4O-MINI, Attack Variant=+ Cipher2025.05 | 26 | 2.63 | 22 | |
| TRIDENT-COREModel=GPT-3.5, Attack Variant=Standard2025.05 | 28 | 2.13 | 5 | |
| TRIDENT-COREModel=GPT-4O-MINI, Attack Variant=+ Low Resource2025.05 | 28 | 2.21 | 9 | |
| TRIDENT-COREModel=GPT-4O-MINI, Attack Variant=+ Past Tense2025.05 | 28 | 2.44 | 14 | |
| TRIDENT-COREModel=GPT-4O-MINI, Attack Variant=+ Persona Modulation2025.05 | 38 | 2.53 | 18 | |
| TRIDENT-COREModel=Mistral-7B, Attack Variant=+ RENELLM2025.05 | 43 | 2.08 | 5 | |
| TRIDENT-COREModel=Qwen-2.5-7B, Attack Variant=Instructions2025.05 | 44 | 4.03 | 64 | |
| TRIDENT-COREModel=Llama-3.0-8B, Attack Variant=+ Persona Modulation2025.05 | 46 | 2.05 | 8 | |
| TRIDENT-COREModel=Qwen-2.5-7B, Attack Variant=Standard2025.05 | 46 | 3.26 | 32 | |
| TRIDENT-COREModel=Llama-3.0-8B, Attack Variant=Standard2025.05 | 49 | 2.32 | 19 | |
| TRIDENT-COREModel=Mistral-7B, Attack Variant=+ Low Resource2025.05 | 56 | 2.26 | 11 | |
| TRIDENT-COREModel=Llama-3.0-8B, Attack Variant=+ Cipher2025.05 | 70 | 3.02 | 33 | |
| TRIDENT-COREModel=Llama-3.0-8B, Attack Variant=+ Past Tense2025.05 | 71 | 3.13 | 42 | |
| TRIDENT-COREModel=Llama-3.2-3B, Attack Variant=+ RENELLM2025.05 | 73 | 2.63 | 23 | |
| TRIDENT-COREModel=Mistral-7B, Attack Variant=+ Persona Modulation2025.05 | 76 | 2.97 | 36 | |
| TRIDENT-COREModel=Llama-3.1-8B-chat, Attack Variant=Instructions2025.05 | 77 | 2.19 | 8 | |
| TRIDENT-COREModel=Llama-3.0-8B, Attack Variant=+ Low Resource2025.05 | 78 | 2.92 | 31 | |
| TRIDENT-COREModel=Llama-3.0-8B, Attack Variant=Instructions2025.05 | 81 | 2.99 | 31 | |
| TRIDENT-COREModel=Llama-3.2-3B, Attack Variant=+ Past Tense2025.05 | 89 | 3.15 | 38 | |
| TRIDENT-COREModel=Llama-3.2-3B, Attack Variant=+ Cipher2025.05 | 92 | 3.62 | 43 | |
| TRIDENT-COREModel=Llama-3.2-3B, Attack Variant=+ Persona Modulation2025.05 | 92 | 3.4 | 37 | |
| TRIDENT-COREModel=Mistral-7B, Attack Variant=+ Cipher2025.05 | 92 | 3.62 | 43 | |
| TRIDENT-COREModel=Llama-3.0-8B, Attack Variant=+ RENELLM2025.05 | 94 | 3.36 | 44 | |
| TRIDENT-COREModel=Llama-3.2-3B, Attack Variant=+ Low Resource2025.05 | 94 | 3.36 | 44 | |
| TRIDENT-COREModel=Llama-3.2-3B, Attack Variant=Standard2025.05 | 95 | 4.42 | 83 | |
| TRIDENT-COREModel=Mistral-7B, Attack Variant=+ Code Injection2025.05 | 96 | 2.56 | 27 | |
| TRIDENT-COREModel=Mistral-7B, Attack Variant=Standard2025.05 | 97 | 4.34 | 83 | |
| TRIDENT-COREModel=Mistral-7B, Attack Variant=Instructions2025.05 | 99 | 3.92 | 55 | |
| TRIDENT-COREModel=GPT-3.5, Attack Variant=Instructions2025.05 | 99 | 3.13 | 36 | |
| TRIDENT-COREModel=Llama-3.0-8B, Attack Variant=+ Code Injection2025.05 | 100 | 3.17 | 35 | |
| TRIDENT-COREModel=Llama-3.2-3B, Attack Variant=Instructions2025.05 | 100 | 3.17 | 35 | |
| TRIDENT-COREModel=Llama-3.2-3B, Attack Variant=+ Code Injection2025.05 | 100 | 2.82 | 29 | |
| TRIDENT-COREModel=Mistral-7B, Attack Variant=+ Past Tense2025.05 | 100 | 2.82 | 29 |