Adversarial Humanities Benchmark - Indiscriminate Weapons: leaderboard
Adversarial Humanities Benchmark restricted to the AILuminate 'Indiscriminate Weapons' 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 | 0 | #161 |
| 2 | Claude Opus 4.6 | 0 | #102 |
| 3 | Claude Haiku 4.5 | 8.9 | #568 |
| 4 | MiniMax-M2.7 | 18.6 | #518 |
| 5 | GPT-5.4 | 24.4 | #123 |
| 6 | Qwen 3.5 27B | 26.7 | #359 |
| 7 | Qwen 3.6 Plus | 26.7 | #222 |
| 8 | Qwen 3.5 Plus (02-15) | 28.9 | #370 |
| 9 | Qwen 3.5 Flash (02-23) | 31.1 | #467 |
| 10 | GPT-5.4 Mini | 33.3 | #389 |
| 11 | Grok 4.1 Fast | 33.3 | #430 |
| 12 | GPT-OSS-120B | 60 | #700 |
| 13 | Gemini 3.1 Pro (Preview) | 61.4 | #94 |
| 14 | Seed 2.0 Mini | 68.8 | #508 |
| 15 | Seed 2.0 Lite | 71.1 | #289 |
Interactive version: theaggregate.ai/benchmark?slug=adversarial-humanities-benchmark-indiscriminate-weapons · How It Works · Data refreshed daily, snapshot 2026-11-13.