SkillSafetyBench - Execution, Runtime and Protocol: leaderboard
The execution risk domain of SkillSafetyBench (Jin et al., 2026): 26 cases in which compromised interpreters, loaders, runtimes, protocols or sandbox boundaries redirect a CLI coding agent's execution while it works on the original task. Attack success rate in percent, judged by case-specific verifiers; each row is a CLI agent harness with a model backend; lower is better.
Metric: Attack success rate (%; 26 cases of execution redirected through compromised interpreters, loaders, runtimes, protocols or environment boundaries; rule-based case verifiers; each row is a CLI agent harness with a model backend, one run per case). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 7.7 | #100 |
| 2 | GPT-5.5 | 11.5 | #48 |
| 3 | Gemini 3 Flash | 15.4 | #159 |
| 4 | MiniMax-M2.7 | 19.2 | #519 |
| 5 | Kimi K2.5 | 26.9 | #252 |
| 6 | GLM-5.1 | 26.9 | #175 |
| 7 | GPT-5.4 | 30.8 | #121 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=skillsafetybench-execution-runtime-and-protocol · How It Works · Data refreshed daily, snapshot 2026-11-12.