HELM AIR-Bench 2024 - #41-42.2: Unauthorized generation - Health data: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-4 risk category "#41-42.2: Unauthorized generation - Health data", under level-3 "#41-42: Privacy Violations/Sensitive Data Combinations" and level-2 "Privacy". The category holds only 15 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: Around February 2027. 87 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 93.33 | #302 |
| 2 | Qwen 3 Next 80B A3B (Thinking) | 93.33 | #312 |
| 3 | Claude Haiku 4.5 (20251001) | 90 | #264 |
| 4 | GPT-OSS-20B | 86.67 | #523 |
| 5 | GPT-5 Nano | 80 | #439 |
| 6 | GPT-OSS-120B | 73.33 | #346 |
| 7 | Claude Sonnet 4 (20250514) | 73.33 | #223 |
| 8 | Claude Sonnet 4.5 | 66.67 | #148 |
| 9 | O3 (2025-04-16) | 63.33 | #125 |
| 10 | Gemini 2.0 Flash | 60 | #351 |
| 11 | O1 (2024-12-17) | 60 | #157 |
| 12 | Yi Chat 34B | 60 | #950 |
| 13 | Claude Opus 4 (20250514) | 60 | #164 |
| 14 | GPT-5 | 53.33 | #105 |
| 15 | Claude 3.7 Sonnet (20250219) | 53.33 | #209 |
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-2024-41-42-2-unauthorized-generation-health-data · How It Works · Data refreshed daily, snapshot 2026-11-05.