NUWA 风险治理 · 研究
Async control: Stress-testing asynchronous control measures for LLM agents
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联系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
事后集成监控在有界红蓝对抗中表现较好，部署风险仍依赖系统设计。
研究问题
异步审核能否足够及时发现破坏，从而支持可信的部署风险论证？
研究设计与假设
受提示限制的攻击策略与迭代监控在多个软件环境竞争，包含保留和分布外测试；另建模型向假想部署外推。
来源发现
在所述校准下，最强集成优于最佳单监控。
延迟伤害生效及要求多次成功动作，可能比单独提高监控准确性更改善建模安全。
NUWA 解读
NUWA 解读：实证检测和模拟部署安全是不同证据层。
局限与相反证据
假设包括受测攻击策略、可读推理访问、无串谋或监控越狱及假想部署；隐蔽的分布外破坏更困难。
研究联系
NUWA 关联：实际监控挑战形式化延迟与恢复要求；BashArena 采用不同同步指标，不能提供匹配发生率。
研究与决策意义
NUWA 建议：保留红队限制与校准数据，另对部署假设和恢复延迟做压力测试。
精确来源位置
- Async control: Stress-testing asynchronous control measures for LLM agents · 2 SettingsSelected original resource https://arxiv.org/html/2512.13526; sha256:1b1574067ef8c25c9f507be0454732520754e71c9624fc7b27a3e9aeabb978bc; archived source capture; exact capture time not independently established
- Async control: Stress-testing asynchronous control measures for LLM agents · 3 Red-Blue Team GameSelected original resource https://arxiv.org/html/2512.13526; sha256:1b1574067ef8c25c9f507be0454732520754e71c9624fc7b27a3e9aeabb978bc; archived source capture; exact capture time not independently established
- Async control: Stress-testing asynchronous control measures for LLM agents · 4 ResultsSelected original resource https://arxiv.org/html/2512.13526; sha256:1b1574067ef8c25c9f507be0454732520754e71c9624fc7b27a3e9aeabb978bc; archived source capture; exact capture time not independently established
- Async control: Stress-testing asynchronous control measures for LLM agents · 5 Estimating risk across a deploymentSelected original resource https://arxiv.org/html/2512.13526; sha256:1b1574067ef8c25c9f507be0454732520754e71c9624fc7b27a3e9aeabb978bc; archived source capture; exact capture time not independently established
- Async control: Stress-testing asynchronous control measures for LLM agents · 5.2 Model Uncertainty and LimitationsSelected original resource https://arxiv.org/html/2512.13526; sha256:1b1574067ef8c25c9f507be0454732520754e71c9624fc7b27a3e9aeabb978bc; archived source capture; exact capture time not independently established
- Async control: Stress-testing asynchronous control measures for LLM agents · 6 DiscussionSelected original resource https://arxiv.org/html/2512.13526; sha256:1b1574067ef8c25c9f507be0454732520754e71c9624fc7b27a3e9aeabb978bc; 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{async-control,
title = {Async control: Stress-testing asynchronous control measures for LLM agents},
url = {https://www.aisi.gov.uk/research/async-control-stress-testing-asynchronous-control-measures-for-llm-agents},
year = {2025},
date = {2025-12-15},
version = {Source capture 2026-10-25},
note = {Source capture 2026-10-25}
}
NUWA 记录 · BibTeX
@misc{async-control-nuwa,
title = {Async control: Stress-testing asynchronous control measures for LLM agents — NUWA research record},
url = {https://whitzard.tech/nuwa/whitzard-index/work/async-control/},
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--async-control.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--async-control.json.}
}
证据快照 2026-10-29.1 · 索引更新 · RSS
参考文献
- [1] Async control: Stress-testing asynchronous control measures for LLM agents. 2025-12-15 ; Official publication masthead and article byline; 2 Settings; 3 Red-Blue Team Game
- [2] Read the full paper