FORTRESS

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

防护如何在国家公共安全限制与良性访问间权衡？

任务与样本

500项专家对抗提示各配良性版本和4–7项二元量表，覆盖三个大领域。

判分与失效情形

逐题二元量表的平均风险分及过拒分，不是实际伤害概率。

系统条件

500对对抗良性提示，单轮，判分实现另须审计。

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： Claude 3.5 Sonnet所报ARS/ORS为14.09/21.80，Gemini 2.5 Pro为66.29/1.40。

效度与解释

固定单轮提示、专业深度及小领域覆盖有限（§6），判官正确性还需单独审计。

适用研究问题

模型建议：同时保留ARS/ORS，并区分原论文估计与后续判分审计。

• FORTRESS: Frontier Risk Evaluation for National Security and Public Safety · AbstractSelected original resource https://arxiv.org/html/2506.14922; sha256:6d8ae92cd9c9950fafbc23990e3a18f67cf4f527869648b8351d21e52bc62efc; archived source capture; exact capture time not independently established
• FORTRESS: Frontier Risk Evaluation for National Security and Public Safety · §5.1 Main ResultsSelected original resource https://arxiv.org/html/2506.14922; sha256:6d8ae92cd9c9950fafbc23990e3a18f67cf4f527869648b8351d21e52bc62efc; archived source capture; exact capture time not independently established
• FORTRESS: Frontier Risk Evaluation for National Security and Public Safety · §6 LimitationsSelected original resource https://arxiv.org/html/2506.14922; sha256:6d8ae92cd9c9950fafbc23990e3a18f67cf4f527869648b8351d21e52bc62efc; archived source capture; exact capture time not independently established