Adversarial Humanities Benchmark - Specialized Advice (Health): leaderboard
Adversarial Humanities Benchmark restricted to the AILuminate 'Specialized Advice (Health)' 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 Opus 4.6 | 0 | #102 |
| 2 | Claude Sonnet 4.6 | 2.1 | #161 |
| 3 | GPT-5.4 Mini | 20.8 | #389 |
| 4 | GPT-5.4 | 26.1 | #123 |
| 5 | Claude Haiku 4.5 | 26.1 | #568 |
| 6 | MiniMax-M2.7 | 26.1 | #518 |
| 7 | Qwen 3.5 27B | 31.2 | #359 |
| 8 | Qwen 3.5 Plus (02-15) | 31.9 | #370 |
| 9 | Qwen 3.6 Plus | 37 | #222 |
| 10 | Qwen 3.5 Flash (02-23) | 42.6 | #467 |
| 11 | GPT-OSS-120B | 44.7 | #700 |
| 12 | Llama 4 Scout Instruct | 47.8 | #1543 |
| 13 | Llama 4 Maverick Instruct | 50 | #1133 |
| 14 | Grok 4.1 Fast | 52.2 | #430 |
| 15 | Qwen 3 Max (Thinking) | 53.2 | #403 (Qwen 3 Max) |
Interactive version: theaggregate.ai/benchmark?slug=adversarial-humanities-benchmark-specialized-advice-health · How It Works · Data refreshed daily, snapshot 2026-11-13.