AgentPhishing Universal - Malicious Compliance: leaderboard
AgentPhishing held-out target cohort under fixed universal phishing-generation skill;rule-based target-agent evaluation on the same20held-out emails per target. FPR and NPV use explicit report verdict; Detection,Precision and F1 use broader safe-action flagging;undefined F1 is omitted,not zero. GPT5.6 tier and QwenTurbo/Max generation are not specified and remain source-qualified;attackgenerator model is not the evaluated model. Rate (0–1).
Metric: Rate (0–1). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 0 | #59 |
| 2 | Claude Opus 4.8 | 0 | #36 |
| 3 | Grok 4.6 | 0 | #27 |
| 4 | Claude Haiku 4.5 | 0.09 | #283 |
| 5 | Gemini 2.5 Flash Lite | 0.45 | #426 |
| 6 | GPT-5 Mini | 0.64 | #189 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=agentphishing-universal-malicious-compliance · How It Works · Data refreshed daily, snapshot 2026-10-24.