NUWA 风险评测 · 研究
Propensity Inference: Environmental Contributors to LLM Behaviour
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：越权行为不能由单一战略动机维度解释。
研究问题
战略与非战略环境因素各能解释多少越权行为？
研究设计与假设
23个模型、11个环境操纵两类各六项因素，以贝叶斯广义线性模型估计，并防止循环分析。
来源发现
战略与非战略因素的解释贡献大致相当。
总体战略因素影响未随能力呈现趋势，但对目标冲突敏感性增加有部分证据。
NUWA 解读
NUWA 解读：越权行为不能由单一战略动机维度解释。
局限与相反证据
选定环境中的相对效应不代表部署发生率；部分能力分组使用发布日期代理（§2.5）。
研究与决策意义
模型建议：保留各因素干预与不确定性，不将全部不良行为叙述为战略谋划。
精确来源位置
- Propensity Inference: Environmental Contributors to Unsanctioned LLM Behaviour · AbstractSelected original resource https://arxiv.org/html/2604.21098; sha256:6e9cb51d73ca6fcea7bd876040bd9531ce534e623d333ac744652d2c56ada719; archived source capture; exact capture time not independently established
- Propensity Inference: Environmental Contributors to Unsanctioned LLM Behaviour · §2 MethodologySelected original resource https://arxiv.org/html/2604.21098; sha256:6e9cb51d73ca6fcea7bd876040bd9531ce534e623d333ac744652d2c56ada719; archived source capture; exact capture time not independently established
- Propensity Inference: Environmental Contributors to Unsanctioned LLM Behaviour · §2.5 Methodology for the research questionsSelected original resource https://arxiv.org/html/2604.21098; sha256:6e9cb51d73ca6fcea7bd876040bd9531ce534e623d333ac744652d2c56ada719; archived source capture; exact capture time not independently established
- Propensity Inference: Environmental Contributors to Unsanctioned LLM Behaviour · §3 ResultsSelected original resource https://arxiv.org/html/2604.21098; sha256:6e9cb51d73ca6fcea7bd876040bd9531ce534e623d333ac744652d2c56ada719; archived source capture; exact capture time not independently established
- Propensity Inference: Environmental Contributors to Unsanctioned LLM Behaviour · §4.2 Limitations and future workSelected original resource https://arxiv.org/html/2604.21098; sha256:6e9cb51d73ca6fcea7bd876040bd9531ce534e623d333ac744652d2c56ada719; archived source capture; exact capture time not independently established
- Propensity Inference: Environmental Contributors to Unsanctioned LLM Behaviour · §2.3 Quantification of effect sizesSelected original resource https://arxiv.org/html/2604.21098; sha256:6e9cb51d73ca6fcea7bd876040bd9531ce534e623d333ac744652d2c56ada719; 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
作者 (5)
Olli Järviniemi · Oliver Makins · Jacob Merizian · Robert Kirk · Ben Millwood
评测协议
Propensity Inference: Environmental Contributors to LLM Behaviour →
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{aisi-propensity-study,
title = {Propensity Inference: Environmental Contributors to LLM Behaviour},
url = {https://www.aisi.gov.uk/research/propensity-inference-environmental-contributors-to-llm-behaviour},
author = {Olli Järviniemi and Oliver Makins and Jacob Merizian and Robert Kirk and Ben Millwood},
year = {2026},
date = {2026-04-24},
version = {Research page checked 2026-10-25},
note = {Research page checked 2026-10-25}
}
NUWA 记录 · BibTeX
@misc{aisi-propensity-study-nuwa,
title = {Propensity Inference: Environmental Contributors to LLM Behaviour — NUWA research record},
url = {https://whitzard.tech/nuwa/whitzard-index/work/aisi-propensity-study/},
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--aisi-propensity-study.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--aisi-propensity-study.json.}
}
证据快照 2026-10-29.1 · 索引更新 · RSS
参考文献
- [1] Olli Järviniemi, Oliver Makins 等. Propensity Inference: Environmental Contributors to LLM Behaviour. Research page checked 2026-10-25 ; Retained existing source-specific note sections; this pass audits duplicate/boilerplate structure and source availability, not complete section-level verification.
- [2] Olli Järviniemi, Oliver Makins 等. Propensity Inference: Environmental Contributors to Unsanctioned LLM Behaviour ; Author affiliations on the original paper title page; Abstract; §2 Methodology