今日速览
| 序号 | 标题 | 来源 | 日期 | 主题 | 推荐等级 |
|---|---|---|---|---|---|
| 1 | Multimodal-SEA: Benchmarking AI Agent Security under Multimodal Social Engineering Attacks | arXiv | 2026-11-05 | RAG | 高 |
| 2 | Optical Network Digital Twin: From Digital Coherent Transmission to Intelligent Network Operations | arXiv | 2026-11-05 | AI-Agent | 高 |
重点论文与技术动态
1. Multimodal-SEA: Benchmarking AI Agent Security under Multimodal Social Engineering Attacks
- 来源：arXiv
- 日期：2026-11-05
- 作者/机构：Zhitao He, Chi Him Zhang, Chengru Han, Yi R Fung
- 主题标签：
RAG,arXiv - 推荐等级：高
- 分类：cs.CL
一句话结论
论文提出交互式基准 Multimodal-SEA，揭示主流多模态智能体面对多模态社会工程攻击均显脆弱，攻击成功率达 7.1% 至 34.5%。
核心内容
- 真实场景基准：基于真实案例构建 608 个专家设计任务，覆盖 14 个领域、152 个攻击场景及中英双语社会文化背景。
- 多模态放大威胁：引入多模态内容使攻击成功率从 19.9% 攀升至 29.2%，即便智能体提高了主动验证覆盖率也难以防范。
- 防护策略存在局限：更严格的安全指令无法提升对欺诈多模态内容的识别能力，且部分模型的低被攻破率伴随着对合法请求的过度拒绝。
方法与数据
- 构建了集成攻击控制与内容生成的红队测试工具套件，在模拟应用中基于工具执行日志对 7 款主流多模态模型开展 6 个维度的评估。
价值判断
- 值得关注：安全指令未能本质解决多模态欺骗识别问题，防御增强会导致合法可用性受损（过度拒绝）。
- 可复用点：Multimodal-SEA 基准任务集及交互式红队评估框架。
- 局限/待核查：被评测的 7 个具体模型及 6 个评估维度的细节在摘要中未明确。
摘要
Personal AI agents face social engineering attacks supported by forged multimodal content, but existing evaluations mainly focus on text-based dialogue, leaving their intrinsic security under such attacks insufficiently studied. We introduce Multimodal-SEA, an interactive benchmark comprising 608 expert-designed tasks grounded in real-world fraud cases, covering 14 domains and 152 attack scenarios with prepared multimodal content. The tasks span four difficulty levels and reflect the distinct social and cultural contexts of Chinese and English environments. Agents inspect content, verify requests, and execute actions in simulated applications, with attack outcomes grounded in verifiable tool execution logs. To enable controlled red teaming of personal agents, we design a multimodal attack harness that integrates task information, attack progression control, and multimodal content construction to evaluate their intrinsic security against adaptive social engineering attacks in simulated environments. Evaluating seven widely used multimodal models across six dimensions reveals that all are vulnerable, with attack success rates of 7.1% to 34.5%. Adding multimodal content increases attack success from 19.9% to 29.2%, despite increased proactive verification coverage. Stronger safety instructions reduce attack success without improving recognition of deceptive multimodal content, while lower attack success rates for some models accompany excessive refusal of legitimate requests.2. Optical Network Digital Twin: From Digital Coherent Transmission to Intelligent Network Operations
- 来源：arXiv
- 日期：2026-11-05
- 作者/机构：Hideki Nishizawa, Enrico Miotto, Andrea Rosso, Takahito Tanimura, Zehao Wang, Mingzhe Han
- 主题标签：
AI-Agent,arXiv - 推荐等级：高
- 分类：eess.SY
一句话结论
本综述系统梳理了从数字相干传输到智能运维的光网络数字孪生（ONDT）技术体系，并提出了基于保真度、动态性和可部署性的统一框架，助力自主光网络构建。
核心内容
- 从生态系统、功能、控制与实现架构四个维度系统组织 ONDT 体系及关键使能技术。
- 提出 FDD（保真度、动态性、可部署性）指标体系，强调按需保真度以兼顾模型的扩展性与泛化性。
- 梳理了物理建模、现网同步与跨域观测等验证实践，探讨了 ONDT 与 AI 智能体的协同演进方向。
方法与数据
- 汇总了基于物理的传输质量（QoT）估计与测量微调技术，涵盖多场景实验与现网验证；具体数据集摘要未明确。
价值判断
- 值得关注：提出了建模关注器件物理效应而非单纯追求精度的“按需保真度”理念，以及 ONDT 与 AI 的协同路径。
- 可复用点：用于指导 ONDT 设计与验证的 FDD 评估框架。
- 局限/待核查：属于全景式综述，未提供具体的量化算法细节，实际落地可行性待现网测试核查。