Indirect Prompt Injection Attack Success Evaluation on Agent Action Goal-Adjacent 2
85IRanyGPT-5.4
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| GPT-5.42026.05 | 85 | 100 | 89.4 | 85.5 | 83.5 | 72.7 | 65 | 71 | |
| GPT-5.52026.05 | 85 | 100 | 97.6 | 81.9 | 77.6 | 69.6 | 67.9 | 66 | |
| Kimi-K2.62026.05 | 84 | 100 | 97.6 | 81.7 | 78.6 | 68.6 | 67 | 66 | |
| Gemini-3.12026.05 | 76 | 100 | 97.4 | 94.6 | 88.2 | 71.9 | 70 | 67 | |
| Sonnet-4.62026.05 | 8 | 100 | 100 | 100 | 75 | 8 | 8 | 6 |