风险评测 · 评测基准

FORTRESS: Frontier Risk Evaluation for National Security and Public Safety

NUWA 研究解读 · 独立复核待完成

首次发表: 2025-06-17来源核验: 2026-10-25

原始来源 ↗引用本记录 ↓化生放核 →网络安全 →

NUWA Research Note

研究问题研究设计与假设来源发现 NUWA 解读局限与相反证据研究与决策意义机构与角色原始资源

模型辅助整理 · 选读原始全文或官方文章中的相关章节；尚未完成逐章节审计。

研究速读

NUWA 解读：限制与良性效用是两个维度，单一拒绝率会掩盖。

研究问题

防护如何在国家公共安全限制与良性访问间权衡？

研究设计与假设

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。

FORTRESS — Figure 1 — Figure 1 · 原文 Figure 1。请结合下方的协议范围与局限阅读。原图与来源 ↗

来源发现

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

这是量表风险与过拒分数，不是伤害概率。

NUWA 解读

NUWA 解读：限制与良性效用是两个维度，单一拒绝率会掩盖。

局限与相反证据

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。

研究与决策意义

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。

精确来源位置

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

来源核查、数据提取、编辑批准与实验复现分别记录。方法说明 →

机构与角色

研究机构 Scale AI

作者 (7)

Christina Q. Knight · Kaustubh Deshpande · Ved Sirdeshmukh · Meher Mankikar · Scale Red Team · SEAL Research Team · Julian Michael

评测协议

FORTRESS →

关联研究

检验其局限 · Auditing FORTRESS: A Multi-Judge Panel Doesn’t Guarantee Robustness

Auditing FORTRESS: A Multi-Judge Panel Doesn’t Guarantee Robustness

审计 FORTRESS 的评分过程如何影响报告分数。

引用与复用

引用研究发现时请引用原始出版物；引用 NUWA 的整理与解释时，单独引用本站记录及对应快照。

BibTeX ↓RIS ↓记录与来源元数据 · JSON ↓阅读覆盖 · JSON 提交更正

原始出版物 · BibTeX

@misc{fortress,
  title = {FORTRESS: Frontier Risk Evaluation for National Security and Public Safety},
  url = {https://arxiv.org/abs/2506.14922},
  author = {Christina Q. Knight and Kaustubh Deshpande and Ved Sirdeshmukh and Meher Mankikar and Scale Red Team and SEAL Research Team and Julian Michael},
  year = {2025},
  date = {2025-06-17},
  version = {arXiv:2506.14922v2},
  note = {arXiv:2506.14922v2}
}

NUWA 记录 · BibTeX

@misc{fortress-nuwa,
  title = {FORTRESS: Frontier Risk Evaluation for National Security and Public Safety — NUWA research record},
  url = {https://whitzard.tech/nuwa/whitzard-index/work/fortress/},
  version = {2026-10-29.1},
  organization = {NUWA Frontier AI Safety Lab},
  howpublished = {Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--fortress.json},
  note = {NUWA research note. Source-reading scope and independent review status are stated on the record. Snapshot 2026-10-29.1. Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--fortress.json.}
}

证据快照 2026-10-29.1 · 索引更新 · RSS

参考文献

[1] Christina Q. Knight ∗, Kaustubh Deshpande ⋄ 等. FORTRESS: Frontier Risk Evaluation for National Security and Public Safety. 2025-06-17. v2 [arXiv]; Author affiliations on the original paper title page; Abstract; §5.1 Main Results
[2] Laurence Wroe. Auditing FORTRESS: A Multi-Judge Panel Doesn’t Guarantee Robustness ; Research background and discussion in the original publication