NUWA 风险评测 · 研究方法
A Decision-Theoretic Formalisation of Steganography With Applications to LLM Monitoring
NUWA 研究解读 · 独立复核待完成
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：差距相对于所选智能体、任务与效用，并非文本绝对属性。
研究问题
不知道干净语言分布时，能否测量隐藏信息？
研究设计与假设
框架在审计者指定任务比较接收者和哨兵的可用信息效用，定义隐写差距。
来源发现
来源提出基于可观察决策而非参考分布的测量。
实验显示哨兵模型族及处理输入意愿会改变差距。
NUWA 解读
NUWA 解读：差距相对于所选智能体、任务与效用，并非文本绝对属性。
局限与相反证据
弱或拒绝处理的哨兵可放大差距，理论适用与实证检出应分开。
研究与决策意义
模型建议：解释为隐藏恶意推理前先审查哨兵能力及任务对照。
精确来源位置
- A Decision-Theoretic Formalisation of Steganography With Applications to LLM Monitoring · AbstractSelected original resource https://arxiv.org/html/2602.23163; sha256:00674902495f2bd855834b61e9c9e5bdb499102b364524406aec6da7f0f0c8a2; archived source capture; exact capture time not independently established
- A Decision-Theoretic Formalisation of Steganography With Applications to LLM Monitoring · §4.2 Setup and NotationSelected original resource https://arxiv.org/html/2602.23163; sha256:00674902495f2bd855834b61e9c9e5bdb499102b364524406aec6da7f0f0c8a2; archived source capture; exact capture time not independently established
- A Decision-Theoretic Formalisation of Steganography With Applications to LLM Monitoring · §4.4 DiscussionSelected original resource https://arxiv.org/html/2602.23163; sha256:00674902495f2bd855834b61e9c9e5bdb499102b364524406aec6da7f0f0c8a2; archived source capture; exact capture time not independently established
- A Decision-Theoretic Formalisation of Steganography With Applications to LLM Monitoring · Appendix G.6: sentinel-family resultSelected original resource https://arxiv.org/html/2602.23163; sha256:00674902495f2bd855834b61e9c9e5bdb499102b364524406aec6da7f0f0c8a2; 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
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。
原始出版物 · BibTeX
@misc{steganography-monitoring,
title = {A Decision-Theoretic Formalisation of Steganography With Applications to LLM Monitoring},
url = {https://www.aisi.gov.uk/research/a-decision-theoretic-formalisation-of-steganography-with-applications-to-llm-monitoring},
year = {2026},
date = {2026-04-29},
version = {Source capture 2026-10-25},
note = {Source capture 2026-10-25}
}
NUWA 记录 · BibTeX
@misc{steganography-monitoring-nuwa,
title = {A Decision-Theoretic Formalisation of Steganography With Applications to LLM Monitoring — NUWA research record},
url = {https://whitzard.tech/nuwa/whitzard-index/work/steganography-monitoring/},
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--steganography-monitoring.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--steganography-monitoring.json.}
}
证据快照 2026-10-29.1 · 索引更新 · RSS
参考文献
- [1] A Decision-Theoretic Formalisation of Steganography With Applications to LLM Monitoring. 2026-04-29 ; Official publication masthead and article byline; Abstract; §4.2 Setup and Notation
- [2] Read the full paper