Malicious Command Interception on OpenClaw 1.0 (test)
83Defense RateClaude Opus 4.6
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Claude Opus 4.6LLM Backend=Claude Opus 4.6, Operational Mode=Baseline2026.03 | 83 | — | |
| Qwen3 MaxLLM Backend=Qwen3 Max, Operational Mode=Baseline2026.03 | 68.1 | — | |
| GPT 5.3 CodexLLM Backend=GPT 5.3 Codex, Operational Mode=Baseline2026.03 | 48.9 | — | |
| Kimi K2.5LLM Backend=Kimi K2.5, Operational Mode=Baseline2026.03 | 27.7 | — | |
| Gemini 3.1 ProLLM Backend=Gemini 3.1 Pro, Operational Mode=Baseline2026.03 | 23.4 | — | |
| DeepSeek V3.2LLM Backend=DeepSeek V3.2, Operational Mode=Baseline2026.03 | 17 | — |