Frontier AI Risk Monitor - Fortress-Privacy/Scams: leaderboard
Scale's FORTRESS adversarial prompts in the privacy and scams subdomain (30 prompts, 5 epochs) graded against per-prompt rubrics; the platform reverses the risk score, so higher is safer.
Metric: Safety score (0-100, higher is safer). Source: airiskmonitor.net. 46 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.7 | 95.05 | #44 |
| 2 | GPT-5.5 | 90.38 | #49 |
| 3 | MiniMax-M2.7 | 88.38 | #518 |
| 4 | Claude Opus 4.5 (Thinking) | 86.67 | #156 (Claude Opus 4.5) |
| 5 | Claude Opus 5.5 | 81.81 | #5 |
| 6 | Claude Opus 4.6 (Thinking) | 80.38 | #102 (Claude Opus 4.6) |
| 7 | MiniMax-M2 | 80 | #674 |
| 8 | Claude Opus 4.8 | 76.29 | #56 |
| 9 | Qwen 3.5 Plus (Thinking) | 75.33 | #232 (Qwen 3.5 Plus) |
| 10 | GPT-5.4 (High) | 74.19 | #123 (GPT-5.4) |
| 11 | GPT-6 Astra | 73.43 | #6 |
| 12 | MiMo-V2.5-Pro | 72.71 | #279 |
| 13 | Grok 4.3 | 72.57 | #333 |
| 14 | GPT-5.6 Sol | 70.67 | #30 |
| 15 | MiMo-V2.6-Pro | 70.19 | #71 |
Interactive version: theaggregate.ai/benchmark?slug=frontier-ai-risk-monitor-fortress-privacy-scams · How It Works · Data refreshed daily, snapshot 2026-11-13.