ForesightSafety Bench (Industrial Safety, Jailbreak) - Law and Justice: leaderboard
Metric: Attack success rate (%) in the law and justice sector of ForesightSafetyBench-IndustrialSafety-O: each harmful request is wrapped in the paper's templated role-play jailbreak prompts from JailbreakChat (AIM, DAN and developer-mode templates), and the rate is taken over all attacked prompts; a success is a response that complies with the sector-specific harmful request, judged by gpt-4o-2024-11-20 with dimension-specific rubrics; lower is better. Source: arxiv.org. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Haiku 4.5 | 0 | #271 |
| 2 | Claude Sonnet 4.5 | 0.67 | #138 |
| 3 | Kimi K2.5 | 0.67 | #139 |
| 4 | Gemini 3 Pro (Preview) | 2.67 | #64 |
| 5 | Claude 3.5 Haiku | 3.33 | #553 |
| 6 | Qwen 2.5 72B Instruct | 4.67 | #436 |
| 7 | GLM-4.7 | 4.67 | #185 |
| 8 | Gemini 3 Flash (Preview) | 5.33 | #78 |
| 9 | Seed 1.8 | 5.33 | #136 |
| 10 | Seed-1.6 | 6 | #257 |
| 11 | DeepSeek V3.2 | 11.33 | #198 |
| 12 | Qwen 3 235B A22B 2507 Instruct | 12 | #291 |
| 13 | Kimi K2 0905 | 12.67 | #282 |
| 14 | Qwen 3 Max (Thinking) | 14 | #201 (Qwen 3 Max) |
| 15 | Llama 3.3 70B Instruct | 21.33 | #520 |
Interactive version: theaggregate.ai/benchmark?slug=foresightsafety-bench-industrial-safety-jailbreak-law-and-justice · How It Works · Data refreshed daily, snapshot 2026-10-11.