HELM AIR-Bench 2024 - #41-42.13: Unauthorized disclosure - Demographic data: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-4 risk category "#41-42.13: Unauthorized disclosure - Demographic data", under level-3 "#41-42: Privacy Violations/Sensitive Data Combinations" and level-2 "Privacy". The category holds only 24 prompts, so scores move in coarse steps. A GPT-4o judge grades each response from compliance (0) to refusal (1); the mean is shown as 0-100, higher meaning more refusals (safer).
Metric: Refusal Rate (%). Source: crfm.stanford.edu. Saturation forecast: Not forecast. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3 Haiku (20240307) | 100 | #780 |
| 2 | GPT-OSS-120B | 95.83 | #346 |
| 3 | Gemini 1.5 Flash (002) | 95.83 | #587 |
| 4 | GPT-OSS-20B | 95.83 | #523 |
| 5 | Grok 3 Mini Beta | 95.83 | #305 |
| 6 | Claude 3.5 Sonnet (20241022) | 93.75 | #302 |
| 7 | Gemini 1.5 Pro (002) | 93.75 | #368 |
| 8 | Claude Haiku 4.5 (20251001) | 93.75 | #264 |
| 9 | Claude 3 Opus (20240229) | 93.75 | #428 |
| 10 | Granite 4.0 H Small | 93.75 | #666 |
| 11 | Claude Sonnet 4.5 | 91.67 | #148 |
| 12 | Gemini 2.0 Flash | 91.67 | #351 |
| 13 | Gemini 3 Pro (Preview) | 91.67 | #71 |
| 14 | Claude 3.5 Sonnet (20240620) | 91.67 | #308 |
| 15 | Qwen 3 Next 80B A3B (Thinking) | 91.67 | #312 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-air-bench-2024-41-42-13-unauthorized-disclosure-demographic-data · How It Works · Data refreshed daily, snapshot 2026-11-05.