Adversarial Frontiers: Minimum-Norm Attack Ensembles for Robustness Evaluation
本文针对当前对抗鲁棒性评估方法的局限性，提出了一种新的评估框架。现有方法通常使用预定义的攻击集成（如AutoAttack）在单一扰动预算ε和少量范数下进行评估，存在三个根本问题：1) 模型间的鲁棒性-扰动曲线可能相交或以不同速率衰减，导致单
2026-07-26 归档
今日共收录 1024 条安全情报，包含 289 个 CVE，422 篇安全通告，以及 50 篇研究论文。
← 返回归档列表Themis Consensus Extension v1: MEV Mitigation by Randomized Delayed Execution and Intent-Hiding Transactions in Application-Specific Blockchains
该论文介绍了 Themis Consensus Extension v1，这是 Mangata 于 2021 年首次发布的一种共识扩展机制，旨在减轻应用特定区块链中的最大可提取价值（MEV）。MEV 源于特权参与者（如验证者）通过选择、排除
Agree on the Model, Verify the Inference: GKR Protocols for HND-Based Transformer Inference
随着大型语言模型（LLM）和Transformer模型在云服务中的广泛部署，用户将推理任务外包给计算服务商成为一种常见模式。然而，这种外包模式引入了信任问题：服务商可能替换模型（例如使用性能较差的版本）或未完整执行推理计算，从而损害用户利益
Risk-Limiting Audits for Parliamentary Majorities
本文针对议会选举中的风险限制审计（RLA）问题提出新方法。传统RLA通常逐席位验证，但政治关注焦点是某党是否赢得议会多数席位，而非每个席位是否准确。作者将多数席位认证视为部分联合检验问题：只需验证报告获胜的政党实际上赢得了至少多数席位。基于
Beyond Heavy Log Curation: Perplexity-Based APT Detection via Unsupervised, Context-Augmented Language Models
本论文提出了一种名为 CAPTAIN（Context-Augmented Perplexity-based Threat Activity log detectIoN）的 APT 检测方法。研究背景是高级持续性威胁（APT）检测的难点在于大
Classical Acceptance Is Not Hybrid Authentication: Measuring X.509 Verifier Semantics in Post-Quantum Migration
本文研究了后量子密码迁移过程中混合X.509证书验证的语义缺陷。混合证书同时携带经典（如RSA、ECDSA）和后量子（如CRYSTALS-Dilithium、Falcon）凭据，其设计为兼容性，通常将后量子证据放置在经典路径验证可能忽略的位
Leaky Language Models: Stealing Architecture and Inference Optimizations via Per-Token Timing
该论文提出了一种名为 LeakyLMs 的攻击方法，通过分析语言模型生成每个 token 的耗时（per-token timing），从远程 API 接口中窃取模型架构和推理优化部署细节，无需直接访问模型内部结构。攻击分为两部分：第一部分针
Enhancing Attack Detection Capabilities in BACnet/IP Networks Using Machine-Learning Models
建筑自动化系统（BAS）通过BACnet/IP等协议管理关键建筑功能，但防御者缺乏检测BACnet特定攻击的工具和标注数据集。本文针对此空白做出三项贡献：首先，修改了CISA的Zeek BACnet解析器，生成统一的逐包日志，简化了机器学习
Deepfake News Detection: A Multimodal Framework Integrating LipNet, DeepSpeech and ResNET for Enhanced Audio-Visual Analysis
本文提出了一种多模态深度伪造新闻检测框架，旨在通过同时分析视频中的音频和视觉线索来鉴别视频内容的真实性。该框架首先从唇部运动（使用LipNet编码）、音频内容（使用DeepSpeech2编码）和视频帧（通过BlazeFace检测人脸并用Re
JANUS: Foreseeing Latent Risk for Long-Horizon Agent Safety
本文针对长周期工具型AI代理的安全问题，提出了一种前瞻性防护框架Janus。传统的内容过滤方法难以应对代理执行复杂任务时因延迟风险导致的故障。Janus通过多智能体模拟合成多样化的代理轨迹，并训练一个共享策略，该策略包含两个耦合任务：1）预
Unconditional Unclonable Encryption
本文提出了一种信息论安全的、一次性的私钥不可克隆加密方案，用于加密单比特消息，且该方案无需依赖任何计算假设或量子硬件。不可克隆加密是一种具有特殊安全性的加密原语：即使攻击者获得了量子层面的加密密文副本，也无法产生两个独立的、能够同时破解密文
ImageMagick before 7.1.2-27 contains a memory leak vulnerability in the magick command-line interface when invalid options are provided. Attackers can trigger memory exhaustion ...
CVE-2026-66011
CVE-2026-66011 是 ImageMagick 7.1.2-27 之前版本中存在的一个内存泄漏漏洞。该漏洞位于 magick 命令行界面中，当用户提供无效选项时，程序未能正确释放已分配的内存，导致内存泄漏。攻击者可以通过反复提交格
In the Linux kernel, the following vulnerability has been resolved: drm/hyperv: validate VMBus packet size in receive callback hyperv_receive_sub() reads msg->vid_hdr.type and...
CVE-2026-64527
Linux 内核 drm/hyperv 驱动中发现一个未验证 VMBus 数据包大小的漏洞。在 hyperv_receive_sub() 回调函数中，驱动程序在处理接收到的 VMBus 消息时，未先检查消息的实际长度。它直接读取消息头中的
In the Linux kernel, the following vulnerability has been resolved: net/handshake: Take a long-lived file reference at submit handshake_nl_accept_doit() needs the file pointer...
CVE-2026-64523
CVE-2026-64523 是 Linux 内核 net/handshake 子系统中的一个漏洞。该漏洞源于文件引用生命周期管理不当：在处理 TLS 握手请求时，`handshake_nl_accept_doit()` 函数需要访问 `r
In the Linux kernel, the following vulnerability has been resolved: net/mlx5e: Fix eswitch mode block underflow on IPsec acquire SA mlx5e_xfrm_add_state() handles acquire-flow...
CVE-2026-64522
该漏洞存在于 Linux 内核的 net/mlx5e 驱动中，涉及 IPsec 安全关联（SA）的 acquire 流程。当 mlx5e_xfrm_add_state() 处理 acquire 流临时 SA 时，它会分配软件状态并跳过硬件卸
In the Linux kernel, the following vulnerability has been resolved: drm/xe/gsc: Fix double-free of managed BO in error path The error path in xe_gsc_init_post_hwconfig() expli...
CVE-2026-64517
In the Linux kernel, the following vulnerability has been resolved: drm/xe/gsc: Fix double-free of managed BO in error
In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu/vce1: Fix VCE 1 firmware size and offsets The VCPU BO contains the actual FW at an offset, but i...
CVE-2026-64516
In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu/vce1: Fix VCE 1 firmware size and offset
In the Linux kernel, the following vulnerability has been resolved: wifi: mac80211: fix MLE defragmentation If either reconf or EPCS multi-link element (MLE) is contained in a...
CVE-2026-64515
In the Linux kernel, the following vulnerability has been resolved: wifi: mac80211: fix MLE defragmentation If either
In the Linux kernel, the following vulnerability has been resolved: userfaultfd: gate must_wait writability check on pte_present() userfaultfd_must_wait() and userfaultfd_huge...
CVE-2026-64514
In the Linux kernel, the following vulnerability has been resolved: userfaultfd: gate must_wait writability check on pt
In the Linux kernel, the following vulnerability has been resolved: KVM: x86: Unconditionally recompute CR8 intercept on PPR update The TPR_THRESHOLD field in the VMCS is used...
CVE-2026-64513
In the Linux kernel, the following vulnerability has been resolved: KVM: x86: Unconditionally recompute CR8 intercept o