HELM AIR-Bench 2024 - #4: Automated Decision-Making: leaderboard
HELM AIR-Bench 2024 (release v1.19.0) refusal rate for the level-3 risk category "#4: Automated Decision-Making", under the level-2 category Operational Misuses of the AIR 2024 regulation-derived taxonomy. Every prompt is a harmful request; a GPT-4o judge grades each response from full compliance (0) to refusal (1), and the mean is shown as 0-100, where higher means 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 Haiku 4.5 (20251001) | 81 | #264 |
| 2 | Claude Sonnet 4 (20250514) | 73 | #223 |
| 3 | GPT-OSS-120B | 71.67 | #346 |
| 4 | Claude Opus 4 (20250514) | 70.67 | #164 |
| 5 | Claude Sonnet 4.5 | 68.33 | #148 |
| 6 | Claude 3.5 Sonnet (20241022) | 66.67 | #302 |
| 7 | Qwen 3 Next 80B A3B (Thinking) | 66.67 | #312 |
| 8 | Claude 3.7 Sonnet (20250219) | 66.33 | #209 |
| 9 | Gemini 1.5 Flash (002) | 65 | #587 |
| 10 | Claude 3 Sonnet (20240229) | 65 | #694 |
| 11 | O3 (2025-04-16) | 62 | #125 |
| 12 | Claude 3.5 Sonnet (20240620) | 59.33 | #308 |
| 13 | GPT-5.1 (2025-11-13) | 59.33 | #107 |
| 14 | GPT-5 | 58 | #105 |
| 15 | Claude 3 Opus (20240229) | 57.33 | #428 |
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-2024-4-automated-decision-making · How It Works · Data refreshed daily, snapshot 2026-11-05.