Adversarial Attack Evaluation on Mind2Web 600 tasks (test)
46.2ASR (Finance, pass@10)Genesis
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GenesisBackend LLM=GPT-4o2025.10 | 46.2 | 47.5 | 47 | 49 | 47.4 | |
| GenesisBackend LLM=GPT-4o, Initialization Setting=w/o Initialization2025.10 | 37.8 | 38 | 38.4 | 38.6 | 38.2 | |
| GenesisBackend LLM=Gemini-2.5-Flash2025.10 | 37.1 | 38.3 | 37.6 | 39.3 | 38.1 | |
| AdvAgentBackend LLM=GPT-4o2025.10 | 37 | 35.9 | 36.8 | 36.6 | 36.6 | |
| GenesisBackend LLM=Gemini-2.5-Flash, Initialization Setting=w/o Initialization2025.10 | 30.4 | 30.6 | 30.7 | 31.1 | 30.7 | |
| AdvAgentBackend LLM=Gemini-2.5-Flash2025.10 | 30.2 | 28.2 | 29.9 | 28.8 | 29.3 | |
| EIABackend LLM=GPT-4o2025.10 | 27.9 | 30.2 | 31.1 | 29.5 | 29.7 | |
| EIABackend LLM=Gemini-2.5-Flash2025.10 | 22.3 | 24.2 | 25 | 23.6 | 23.8 | |
| InjecAttackBackend LLM=GPT-4o2025.10 | 20.5 | 20.8 | 21.3 | 23 | 21.4 | |
| GenesisBackend LLM=GPT-52025.10 | 19.7 | 20.5 | 20 | 21.1 | 20.3 | |
| InjecAttackBackend LLM=Gemini-2.5-Flash2025.10 | 16.4 | 16.6 | 17 | 18.4 | 17.1 | |
| AdvAgentBackend LLM=GPT-52025.10 | 16 | 15 | 16 | 15.3 | 15.6 | |
| GenesisBackend LLM=GPT-5, Initialization Setting=w/o Initialization2025.10 | 15.8 | 16.8 | 16.1 | 16.8 | 16.4 | |
| AgentAttackBackend LLM=GPT-4o2025.10 | 13.2 | 12.8 | 14.5 | 13.1 | 13.4 | |
| EIABackend LLM=GPT-52025.10 | 11.9 | 12.9 | 13.4 | 12.5 | 12.7 | |
| AgentAttackBackend LLM=Gemini-2.5-Flash2025.10 | 10.1 | 10.2 | 11.9 | 10.5 | 10.7 | |
| InjecAttackBackend LLM=GPT-52025.10 | 8.8 | 8.8 | 9.1 | 9.8 | 9.1 | |
| AgentAttackBackend LLM=GPT-52025.10 | 5.3 | 5.2 | 6.3 | 5.5 | 5.6 | |
| I-GCGBackend LLM=GPT-4o2025.10 | 2.9 | 2.2 | 3.8 | 5.5 | 3.6 | |
| I-GCGBackend LLM=Gemini-2.5-Flash2025.10 | 2.3 | 1.2 | 3.1 | 3.8 | 2.6 | |
| I-GCGBackend LLM=GPT-52025.10 | 1.1 | 0.5 | 1.9 | 2.4 | 1.5 | |
| GCGBackend LLM=GPT-4o2025.10 | 0.3 | 2 | 1.1 | 2.7 | 1.5 | |
| GCGBackend LLM=Gemini-2.5-Flash2025.10 | 0.2 | 1.9 | 0.5 | 1.7 | 1.1 | |
| GCGBackend LLM=GPT-52025.10 | 0.2 | 1.2 | 0.1 | 0.8 | 0.6 |