Computer Use Control-Flow Hijacking on CFH-Hard Computer Use
67Gen. RateUndefended
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Undefended2025.10 | 67 | 83 | 89 | 89 | 78 | |
| ACF2025.10 | 67 | 83 | 89 | 89 | 78 | |
| LP2025.10 | 67 | 83 | 89 | 89 | 78 | |
| LlamaFirewallBackbone=llama2025.10 | 67 | 83 | 89 | 89 | 78 | |
| LlamaFirewallBackbone=4o-mini2025.10 | 67 | 78 | 78 | 78 | 72 | |
| LlamaFirewallBackbone=4o2025.10 | 56 | 72 | 78 | 83 | 72 | |
| LlamaFirewallBackbone=o4-mini2025.10 | 6 | 39 | 72 | 11 | 50 | |
| CONTROLVALVE2025.10 | 0 | 0 | 0 | 0 | 0 |