Red-teaming tool-calling agents on MCP Slack 1.0 (test)
90.6RRZero-Shot (ZS)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Zero-Shot (ZS)Mode=Zero-Shot2026.03 | 90.6 | 0 | |
| Multi-Turn (MT)Mode=Multi-Turn2026.03 | 48.4 | 10.9 | |
| Iterative Refinement (IR)Mode=Iterative Refinement2026.03 | 34.4 | 10.9 | |
| Standard Evolution (SE)Mode=Evolutionary Search2026.03 | 26.6 | 28.1 | |
| T-MAPBackbone=DeepSeek-V3.2, Target Model=GPT-5-mini2026.03 | 15.6 | 64.1 |