SandboxEscapeBench

档案区分来源方法阅读与既有目录信息；已经公开不代表完成独立证据复核或双语编辑复核。

测量构念

受控条件下，智能体能否逃出刻意弱化的容器沙箱？

任务与样本

嵌套隔离设计让有Shell权限的对抗智能体测试内层弱点，外层限制实验影响。

判分与失效情形

逐样本或至少一次逃逸成功须保留重复语义。

系统条件

有Shell的对抗智能体、内层人为弱点及嵌套隔离。

统计含义

来源报告的结果表述；这不等同于已确定运行次数、分母或估计方法： 更新结果报告Opus 4.5在56%样本上成功。 Mythos Preview每个样本至少成功一次，不等于单次运行100%成功（§6.1）。

效度与解释

最少监测、人为弱点与专家估难不代表有防护生产环境的失陷概率（§7）。

适用研究问题

模型建议：区分原始与更新结果，并保留重复次数语义。

• Quantifying Frontier LLM Capabilities for Container Sandbox Escape · AbstractSelected original resource https://arxiv.org/html/2603.02277; sha256:078f48e266792c2b63ee79c3425750429f323b2cf104dacbfa452ce6774da4e7; archived source capture; exact capture time not independently established
• Quantifying Frontier LLM Capabilities for Container Sandbox Escape · §3 MethodologySelected original resource https://arxiv.org/html/2603.02277; sha256:078f48e266792c2b63ee79c3425750429f323b2cf104dacbfa452ce6774da4e7; archived source capture; exact capture time not independently established
• Quantifying Frontier LLM Capabilities for Container Sandbox Escape · §6.1 Primary ResultsSelected original resource https://arxiv.org/html/2603.02277; sha256:078f48e266792c2b63ee79c3425750429f323b2cf104dacbfa452ce6774da4e7; archived source capture; exact capture time not independently established
• Quantifying Frontier LLM Capabilities for Container Sandbox Escape · §7 LimitationsSelected original resource https://arxiv.org/html/2603.02277; sha256:078f48e266792c2b63ee79c3425750429f323b2cf104dacbfa452ce6774da4e7; archived source capture; exact capture time not independently established