FinSafetyBench - Financial Crimes (Chinese): leaderboard
FinSafetyBench (Hou et al., 2026) has 1,881 harmful financial instructions in aligned English and Chinese versions, built from Chinese court judgments on financial crimes, CFA ethics cases and financial items of existing red-teaming sets. This board covers the Chinese instructions of the 11 financial-crime categories (market manipulation, fraud, money laundering and others). Attack success rate in percent, averaged over the PAIR, ReNeLLM and FlipAttack jailbreaks, counting a response only when a Qwen3-32B judge rates it a complete violation (10 of 10); lower is better.
Metric: Attack success rate (%; mean over the PAIR, ReNeLLM and FlipAttack jailbreaks; share of the benchmark's financial-crime instructions in Chinese whose jailbroken response a Qwen3-32B judge rates a complete violation (10 of 10)). Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3 8B Instruct | 52.65 | #4652 |
| 2 | InternLM3-8B-Instruct | 64.96 | #2586 |
| 3 | Qwen 2.5 32B Instruct | 80.73 | #1216 |
| 4 | glm-4-9B (0414) | 87.02 | #2196 |
| 5 | Mistral Small 3 | 91.46 | #1786 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=finsafetybench-financial-crimes-chinese · How It Works · Data refreshed daily, snapshot 2026-11-12.