FinSafetyBench (PAIR) - Ethical Violations (English): leaderboard
Metric: Attack success rate (%; PAIR jailbreak; share of the benchmark's ethical-violation instructions in English whose jailbroken response a Qwen3-32B judge rates a complete violation (10 of 10)). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3 8B Instruct | 45.34 |
| 2 | GLM-4 9B (0414) | 71.31 |
| 3 | Qwen 2.5 32B Instruct | 72.03 |
| 4 | InternLM3-8B-Instruct | 72.83 |
| 5 | Qwen 2.5 72B Instruct | 77.56 |
| 6 | Mistral Small 3 | 81.71 |
Interactive version: theaggregate.ai/benchmark?slug=finsafetybench-pair-ethical-violations-english · How It Works · Data refreshed daily, snapshot 2026-09-26.