MultiGait: A Multi-Sensor Multi-Perspective Multi-Session Biometric Inference Benchmark and its Dataset
MultiGait 论文针对智慧城市新兴传感器（热成像相机、深度相机、激光雷达等）缺乏合适数据集、难以系统评估其隐私风险的问题，构建了首个多传感器、多视角、多会话的步态数据集，并形成了一套身份推断基准。作者认为，在大量关于传感器隐私影响的未
2026-09-03 归档
今日共收录 1407 条安全情报，包含 491 个 CVE，487 篇安全通告，以及 120 篇研究论文。
← 返回归档列表Context Inference Attacks Without Jailbreaks
本文研究 Agentic AI 系统在推理时处理敏感数据（如医疗记录、金融文档）时面临的隐私泄露问题。与以往主要关注通过越狱攻击直接诱导模型泄露敏感内容的研究不同，本文首次系统化地研究了 Agent 自身的工具调用所组装出的隐藏上下文（hi
Cheap Verifiers, Large Blind Spots: Measuring the Reliability Cost of Cost-Saving Cascades
本文研究的是大语言模型（LLM）推理级联（inference cascade）中的可靠性代价问题。推理级联是一种常见的成本优化策略：大部分查询由一个廉价小模型处理，只有困难样本才升级到能力更强的前沿大模型作为验证器。一种自然扩展是闭环自我改
CAPTCHAs in the Agentic Era: Solvers That Learn from Every Encounter
这篇论文研究在智能体（agent）时代，视觉验证码（CAPTCHA）面临的新威胁。作者指出，基于视觉语言模型（VLM）的智能体虽然能解决验证码，但每次都是从零开始，效率低；而专门的检测器虽然快，但只能处理训练过的类别。论文提出一个核心问题：
Removing Speech, Keeping Activities: A Privacy Firewall for Acoustic Sensing in Assisted Living
本文提出了一种面向辅助生活场景的声学传感隐私防火墙，旨在解决环境音频监测中的语音隐私问题。研究背景是：声学传感为非侵入式监测老年人日常活动提供了有前景的手段，但语音隐私顾虑成为实际部署的关键障碍。核心方法是使用基于 U-Net 编码器-解码
Pushing Forward Multi-Secret-Key Homomorphic Encryption for Private Average Aggregation
该论文提出一种面向联邦学习中私有平均聚合的多密钥同态加密协议。联邦学习允许多个客户端在不共享原始数据的情况下训练共享模型，但传输的模型更新仍可能泄露敏感信息，因此私有聚合是实际部署中的关键组件，尤其是跨孤岛场景。传统单密钥同态加密（HE）假
Adversarial Vulnerabilities of Neural Biomarker Identification Systems
本文针对基于脑电图（EEG）信号的神经生物特征认证系统展开安全性研究。现有工作大多关注深度学习分类器在白盒条件（即攻击者完全掌握模型参数与结构）下的对抗攻击，而对其他类型的神经特征识别方法以及更实际的仅具备黑盒查询权限的攻击场景关注不足。为
Sentinel-Based Failover for QKD-Augmented IPsec Tunnels
量子计算的发展威胁现有公钥加密体系，量子安全IPsec成为过渡期的重要方案。业界常将量子密钥分发（QKD）与传统密钥交换结合以增强安全性，但这带来了运维挑战：当QKD基础设施发生故障时，如何维持IPsec隧道的高可用性？本文设计、实现并实验
SoK: Motion Data Privacy in Extended Reality
本文是一篇关于扩展现实（XR）中运动数据隐私的系统化知识综述（SoK）。XR 设备（如 VR/AR 头显）依赖对用户身体动作的持续追踪来实现沉浸式交互，例如抓取、注视或移动虚拟物体。然而，运动追踪数据记录了个体独特的运动模式，现有研究表明这
PhantomCall: Evading ML Malware Detectors via Function Call Graph Perturbation
论文《PhantomCall: Evading ML Malware Detectors via Function Call Graph Perturbation》研究了面向Windows PE恶意软件机器学习检测器的对抗攻击。现有攻击通常
Verifiable quantum advantage in extremely low depth
这篇论文提出了一种可在极低深度量子电路上求解、可被经典计算机高效验证、并且在基于格（Lattice）的假设下对多项式时间经典算法困难的新采样问题。该量子采样器有两种实现方式：一种使用对数-对数深度（log-logarithmic depth
tsi-coop tsi-dpdp-cms client-side enforcement of server-side security
CVE-2026-84841
A security flaw has been discovered in tsi-coop tsi-dpdp-cms up to 0.5.0. This vulnerability affects unknown code. The m
tsi-coop tsi-dpdp-cms Bootstrap Setup Endpoint InterceptingFilter.java missing authentication
CVE-2026-84840
A vulnerability was identified in tsi-coop tsi-dpdp-cms up to 0.5.0. This affects an unknown part of the file Intercepti
tsi-coop tsi-dpdp-cms Admin Console/DPO Compliance Console web.xml missing authentication
CVE-2026-84839
A vulnerability was determined in tsi-coop tsi-dpdp-cms up to 0.5.0. Affected by this issue is some unknown functionalit
HTTPX2: Streaming response decompression does not bound peak memory (decompression amplification)
CVE-2026-84382
HTTPX2 is a next generation HTTP client for Python. Prior to 2.12.0, the HTTPX2 content decoders in src/httpx2/httpx2/_d
HTTPX2: Secure WebSocket traffic sent without TLS through SOCKS proxies
CVE-2026-84381
HTTPX2 is a next generation HTTP client for Python. Prior to 2.10.0, httpcore2 fails to start TLS in src/httpcore2/httpc
HTTPX2: Conflicting Content-Length and Transfer-Encoding headers can be auto-generated
CVE-2026-84380
HTTPX2 is a next generation HTTP client for Python. Prior to 2.11.0, Request._prepare() in src/httpx2/httpx2/_models.py
HTTPX2: Multipart part header injection via unvalidated file Content-Type and custom headers
CVE-2026-84379
HTTPX2 is a next generation HTTP client for Python. Prior to 2.11.0, FileField.render_headers() in src/httpx2/httpx2/_mu
PREVAIL: ALU32 pointer arithmetic accepted without is64 gate — verifier emits false PASS for pointer-corrupting programs
CVE-2026-53706
PREVAIL is a Polynomial-Runtime EBPF Verifier using an Abstract Interpretation Layer. Prior to version 0.2.4, the prevai
PREVAIL: Non-singleton typeset in add() skips offset update, allowing OOB access to pass eBPF verification
CVE-2026-53670
PREVAIL is a Polynomial-Runtime EBPF Verifier using an Abstract Interpretation Layer. Prior to version 0.2.4, in the Pre