Denial of Service (DoS) attack on ToolSafe (test)
61,981Think Lengthbeam-search optimization framework
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| beam-search optimization frameworkTarget Model=Claude-3.5-Haiku, Template=ToolSafe2026.06 | 61,981 | 27.7 | |
| ReasoningBombTarget Model=Claude-3.5-Haiku, Template=ToolSafe2026.06 | 2,379 | 1.2 | |
| RECUR†Target Model=Claude-3.5-Haiku, Template=ToolSafe2026.06 | 2,298 | 1.16 | |
| OverThinkTarget Model=Claude-3.5-Haiku, Template=ToolSafe2026.06 | 2,283 | 1.15 | |
| CRABSTarget Model=Claude-3.5-Haiku, Template=ToolSafe2026.06 | 2,270 | 1.14 | |
| ENGORGIOTarget Model=Claude-3.5-Haiku, Template=ToolSafe2026.06 | 2,207 | 1.11 | |
| ThinkTrap†Target Model=Claude-3.5-Haiku, Template=ToolSafe2026.06 | 2,207 | 1.11 | |
| Length-matched benignTarget Model=Claude-3.5-Haiku, Template=ToolSafe2026.06 | 1,984 | 1 |