Adversarial Humanities Benchmark - Specialized Advice (Elections): leaderboard
Adversarial Humanities Benchmark restricted to the AILuminate 'Specialized Advice (Elections)' hazard: harmful requests of that kind rewritten as humanities tasks; the number is the share answered unsafely by a three-judge majority. Per-hazard cells rest on far fewer prompts than the headline.
Metric: Attack success rate (%; share of humanities-style rewritten harmful requests answered with unsafe content, 3-judge majority vote, lower is better). Source: arxiv.org. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 7.1 | #161 |
| 2 | GPT-5.4 Mini | 26.2 | #389 |
| 3 | Claude Opus 4.6 | 31 | #102 |
| 4 | Claude Haiku 4.5 | 31 | #568 |
| 5 | GPT-5.4 | 42.9 | #123 |
| 6 | Qwen 3.5 27B | 42.9 | #359 |
| 7 | MiniMax-M2.7 | 43.2 | #518 |
| 8 | Qwen 3.6 Plus | 54.8 | #222 |
| 9 | Qwen 3.5 Flash (02-23) | 54.8 | #467 |
| 10 | Qwen 3.5 Plus (02-15) | 54.8 | #370 |
| 11 | Gemma 4 31B (IT) | 71 | #342 |
| 12 | GPT-OSS-120B | 71.1 | #700 |
| 13 | Mistral Small 4 | 73.8 | #1201 |
| 14 | GPT-OSS-20B | 75.6 | #1147 |
| 15 | DeepSeek V3.2 | 80.5 | #406 |
Interactive version: theaggregate.ai/benchmark?slug=adversarial-humanities-benchmark-specialized-advice-elections · How It Works · Data refreshed daily, snapshot 2026-11-13.