- 企业网络攻击应对：国家指南目录
九个国家的程序指南、本地语言版本及不可执行的BPMN响应流程。
- Cyberattack response for companies in the United States
Evidence, reports, counsel, insurance and public communications after a cyberattack in the United States.
- Cyberattack response for companies in the United Kingdom
Evidence, reports, counsel, insurance and public communications after a cyberattack in the United Kingdom.
- Cyberattack response for companies in Canada
Evidence, reports, counsel, insurance and public communications after a cyberattack in Canada.
- Cyberattack response for companies in Australia
Evidence, reports, counsel, insurance and public communications after a cyberattack in Australia.
- Cyberattack response for companies in Germany
Evidence, reports, counsel, insurance and public communications after a cyberattack in Germany.
- Cyberattack response for companies in France
Evidence, reports, counsel, insurance and public communications after a cyberattack in France.
- Cyberattack response for companies in the Netherlands
Evidence, reports, counsel, insurance and public communications after a cyberattack in the Netherlands.
- 新加坡企业遭受网络攻击后的应对流程
面向新加坡企业的网络攻击应对：证据保全、报告、律师协作、保险索赔及公开沟通。
- Cyberattack response for companies in Japan
Evidence, reports, counsel, insurance and public communications after a cyberattack in Japan.
- Cybersecurity daily — 1 November 2026: MCP migration and corrected extension scope
Smithery's connection-management shutdown date has arrived; Cursor's notice also names 2.1.312. Distinguish planned shutdown, observed operation and verification time.
- Cybersecurity daily — 31 October 2026: extension removal is not incident recovery
Cursor extension follow-up, Wasmtime UDP authorization bypass, and updated AWS Brotli mitigation status. Check host privileges, network authorization and deployed patches.
- Cybersecurity daily — 30 October 2026: verify enforcement beyond configuration
Clerk adds passkeys as a second factor; an older PDF MCP authentication flaw enters the advisory database. Verify enforcement, not just settings and logs.
- Cybersecurity daily — 29 October 2026: verify that Brotli patches are active
AWS details Brotli mitigation: phased Lambda rollout, Node.js 24 verification, container rebuilds and upstream-response decoding.
- Cybersecurity daily — 28 October 2026: separate detection, delegation and runtime authorization
Zuma/Agent Gateway runtime authorization and Socure Agent Intelligence: distinguish release stages and authority boundaries.
- Cybersecurity daily — 27 October 2026: the boundary around agent-managed IAM
Google Cloud introduces IAM Helper. Separate diagnostics and simulation from permission changes, and verify the approval-to-execution boundary.
- Why Faster Cybercrime Intake May Not Reduce Case Backlogs
Public Japanese statistics and an explicit queueing model show why faster intake may shift a bottleneck downstream. Evidence preservation, international assistance, and deterrence require distinct measurements.
- Crypto fraud and theft: Japanese requests for Canadian asset restraint and seizure
For Japanese victims and receiving police officers: RCMP cooperation, formal central-authority requests, evidence and translations, procedural-stage conditions and asset return.
- Prompt injection is not SQL injection: the NCSC design lesson
The NCSC blog says prompt injection is not SQL injection: current large language models do not separate instructions from data, and the risk may never be fully removed the way parameterized queries remove SQL injection (https://www.ncsc.gov.uk/blog-post/prompt-injection-is-not-sql-injection). The NCSC news warns that the comparison is dangerous and challenges claims that the issue can be stopped (https://www.ncsc.gov.uk/news/mistaking-ai-vulnerability-could-lead-to-large-scale-breaches). Design lesson only. Not an attack guide and not legal advice.
- 向客户说明欧盟人工智能法第50条标识时该说什么
欧盟委员会的 Article 50 页面分别规定了提供者与部署者在交互告知、机器可读标记和深度伪造披露方面的义务（https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50）。一份律师事务所警示称，这些义务于 2 August 2026 开始适用，并未随高风险规则一并延期（https://www.goodwinlaw.com/en/insights/publications/2026/08/alerts-technology-dpc-eu-ai-act-transparency-obligations-now-in-force）。这是一般信息，不是法律意见。
- 跟随行业工作流的网络安全演练
24 个行业视角、18 个技术面，以及八项离线演练，把 CTF 学习连接到连续性、修复与恢复。
- AI 视频、声音冒充与权限边界
对 Tavus、已报告的 AI 滥用，以及采用者应当测试的身份、同意和工具权限边界所作的有出处综述。
- System One Coding：通过决策来组合代码
带类型的积木、分阶段验证、已测量的比较，以及一项注册制免费试点。
- 账号被盗之后：向警察咨询之前要做的「时间顺序表」和「证据一览」（可复制使用）
账号被盗之后、向警察咨询之前不要删掉的东西。用时间顺序表和证据一览的模板。如果你在日本：窗口（被害申告、咨询、线上）的差别整理在这一页。
- 遭遇加密资产诈骗之后：给警察和交易所带去的「汇款一览表」要怎样做
在加密资产·投资诈骗的受害里，在追加汇款或向「追回业者」行动之前。把汇款和对话整理成 1 张，准备向警察和交易所咨询。如果你在日本：警察咨询电话是 #9110。本页不承诺追回。
- IC3 投诉工作表：提交之前要收集什么（加密资产与网络诈骗）
把 FBI 的 IC3 所询问的每一项都备好——损失、交易哈希、对方的详情、邮件头——然后提交一份完整的投诉，并留存你的副本。
- 企业为何需要一套孪生体，以及 Mithril 如何提供它
生产环境不适合用来弄清谁能触达敏感系统、一次拒绝会是什么样子，以及正在运行的集群是否就是当初声明的那一套。孪生体保留这些名称，并且不应用结果。Mithril 把这套孪生体做成一份文档：以反向域名命名的组件可以组合，一次运行返回响应与判定，集群对照观测进行核验而不被启动。
- 密钥未被盗，一小时内 $351 Million — Bitget 热钱包入侵与审批流的信任边界
Bitget 于 9 月 25 日确认，约 $351.6 million 在 9 月 24 日 18:31 UTC 被发现的未经授权转账中，从其热钱包与温钱包的若干部分流出。据报其离线冷钱包未受影响，提现处于暂停，该交易所表示其 $464 million 的 User Protection Fund 覆盖全部损失。据 Bitget 所述，攻击者并未取走私钥，而是进入了一个后端系统并伪造交易数据，从而把审批流导向放出资金。我们审视 Lazarus 归因问题，以及它与本观察站近几周所论述的管理与身份平面主题之间的同构。
- 身份签发方本身倒下的一周 —— F5 BIG-IP APM OAuth 零日（CVE-2026-94127）与 KEV 在管理面和身份面上的集中
F5 于 9 月 22 日披露，BIG-IP APM 的 OAuth 路径存在一个严重零日（CVE-2026-94127，CVSS 9.8），允许未经认证的远程代码执行，并确认已有在野利用。CISA 于同日将其列入 KEV 目录，联邦期限为 9 月 25 日。Arista VCO 与 2 个 Check Point 管理服务器缺陷也在当天进入 KEV。本文考察针对边界设备管理面与身份面的攻击如何汇合。
- 通往本地 root 的道路在同一周内被一并打开 —— Linux 内核 LPE 列入 KEV 与公开漏洞利用代码的合流
9 月 18 日，CISA 将三个已确认遭受实际攻击的 Linux 内核漏洞（CVE-2025-39682、CVE-2026-53266、CVE-2025-39964）列入 KEV 目录，Red Hat 确认存在公开的漏洞利用代码。同日，四个内核 LPE（DirtyAH6、TUNderflow、PPPoEject、DiagSpill）的可用漏洞利用代码在协调披露下公开发布。本文梳理为何“共享主机的特权提升”正在继“边界设备”之后成为焦点。
- 最大严重度的认证绕过击中“身份判定本体”—— Cisco ISE CVE-2026-76460 与边界设备 root 侵害的重复结构
Cisco ISE / ISE-PIC 的 API 认证绕过 CVE-2026-76460（CVSS 10.0）正处于实际攻击之下。CISA 已于 9 月 16 日将其列入 KEV 目录，联邦机构的修复期限为 9 月 19 日。该漏洞没有 workaround，iACL 仅是缓解措施。本文梳理其与 9 月 14 日公开的 Secure Email Gateway 零日（CVE-2026-76461）在一周内并行的“边界设备 root 侵害”结构，以及被侵害设备自身的日志不再是一手信息这一含义。
- AI 网关成为攻击面 —— LiteLLM 漏洞群列入 KEV，以及通向密钥、MCP 与云端的攻击路径
BerriAI LiteLLM 的 MCP 认证绕过（CVE-2026-59822）已被列入 CISA 的 KEV 目录，Microsoft 报告有攻击者经 LiteLLM 网关分发 XMRig。Wiz 的研究指出，约 3,000 个公开实例中约一成接受默认密钥 sk-1234 或完全无认证。本文梳理 AI 基础设施作为"凭证宝库"的含义。
- 当智能体的"身体"被劫持——一个扩展就能驱动五款产品的 AI 助手
Forever Security 的实证显示，一个仅持有普通权限的浏览器扩展就能驱动五款产品——Chrome、Comet、Edge、Opera Neon 与 Claude——的内置 AI 助手。另一方面，Mandiant 报告了一起攻击者劫持 AI 编码助手会话、把 Shai-Hulud 蠕虫扩散到约 100 个仓库的事件。本文梳理智能体运行时的信任边界。
- 「补丁空窗期」零日漏洞链 —— Chromium 已修复、Chrome 未修复的漏洞在一周内被多个间谍组织共用
Volexity 观测到 BlueMoon 攻击链——Chrome V8 沙箱内任意读写、沙箱逃逸与 Windows ALPC 滥用——被至少多个中国背景攻击集群共用。上游 Chromium 已有修复，但稳定版 Chrome 尚未发布相应更新，形成危险的"补丁空窗期"。本文梳理其影响。
- VC 简报：SYN Ventures — 由运营者主导的网络安全专业基金
一支管理资产为 $900M、由前 CISO 与创始人建立的运营者主导基金。本文将其 2 只基金与 ~40 家公司的组合，对照 mithril.fund 的位置加以梳理。
- Cisco Secure Email Gateway 零日漏洞 CVE-2026-76461 —— 当邮件解析中的 SQL 注入一路通向 root 执行
CVE-2026-76461（CVSS 9.8）正被实际利用，并已加入 CISA 的 KEV 目录。本文梳理反复出现的"边界设备 root 沦陷"攻击模式，以及当设备自身日志不再可信时意味着什么。
- 安全观测博客开通了
每天观测、分析网络安全新闻、重大事件与 AI 模型动态，并在此发布。
- 邮件网关后门写入文件 —— FortiMail 零日漏洞 CVE-2026-104286 与 IBE 功能
Fortinet 于 10 月 1 日披露 FortiMail 路径遍历零日（CVE-2026-104286，CVSS 9.8），允许未经认证的任意文件写入，并确认在野利用。CISA 于同日将其列入 KEV 目录，联邦期限为 10 月 4 日。四个受影响分支均无已发布修复，需先执行 IBE 禁用等临时缓解，本文将其与 9 月下旬 KEV 激增放在一起整理。
- 从诈骗受害者到企业白帽：一幅网络安全图景（2026）
十三个类别，从对诈骗受害者的帮助到企业安全测试，附有已报告的损失、已宣布的融资、按发布方列出的市场预测，以及 Mithril 所处位置的如实状态。
- 基于 SAML 条件的 NetScaler 内存溢出 — CVE-2026-88779 被列入 KEV（截止日期 10 月 7 日）
Citrix 披露了 NetScaler ADC / Gateway 的内存缓冲区边界缺陷（CVE-2026-88779，CVSS v4.0 8.7，CWE-119）。Citrix 称其观察到针对未缓解部署的定向攻击，这些攻击可能导致拒绝服务。CISA 的 KEV 列入表明该漏洞已被利用，其简短说明称该缺陷可能导致拒绝服务。这两处来源都没有确认受影响部署已被成功中断。CISA 于 10 月 4 日将其加入 KEV，联邦机构截止日期为 10 月 7 日。连同 9 月 27 日加入的 CVE-2026-88771 与 CVE-2026-88772，这是两周内加入的第 3 个 NetScaler KEV 条目，并不是该产品有史以来的第 3 个（目录中已有更早条目，包括 CVE-2023-4966）。Global Deny List 签名（v24 或更高版本）是临时缓解措施，修复版本为 14.1-73.41 / 13.1-64.28 或更高版本。
- Unauthenticated API access in Cisco Catalyst SD-WAN Manager — CVE-2026-76504 Added to KEV (Due October 3)
Cisco published CVE-2026-76504 (CVSS 3.1 9.8 per Cisco, CWE-177) in advisory cisco-sa-sdwan-webauth-xr8beuuU: improper URI-encoding handling in the API session-based authentication management of Cisco Catalyst SD-WAN Manager can let an unauthenticated remote attacker reach the API with admin privileges. CISA added it to KEV on September 30, 2026, with a federal due date of October 3. Fixed releases start at 20.9.10.1 / 20.12.8.2 / 20.15.6.1 / 20.18.4.1 / 26.1.2.1 / 26.2.1 (earlier branches migrate). No known workarounds exist; interim mitigation is restricting internet access to trusted hosts, plus the optional Live Protect shield. Cisco has confirmed exploitation in the wild.
- Apple CoreGraphics zero-day added to KEV — CVE-2026-86950, report of exploitation in an extremely sophisticated targeted attack (Due October 2)
On September 28, 2026, Apple released iOS 26.7.1 / iPadOS 26.7.1 / macOS Tahoe 26.7.1 / macOS Sequoia 15.8.1, fixing an out-of-bounds write (CVE-2026-86950, CWE-787) in CoreGraphics. Processing a maliciously crafted file may lead to arbitrary code execution; Apple says it is aware of a report that the issue may have been exploited in an extremely sophisticated attack against specific targeted individuals on iOS versions before iOS 27. CISA added the vulnerability to KEV on September 29, 2026, with a federal due date of October 2 (forensicTriage Yes). The CISA-ADP CVSS 3.1 base is 8.8 (High). The fix is the listed releases only; the primary sources reviewed for this post state no interim workaround.
- Binance: 'No wallet key leaked' - the largest hot-wallet incident on record (10/29) and first observations
From the night of October 29 UTC Binance disclosed a security incident involving unauthorized withdrawals from its hot wallets. Its verified frozen total stood at $2.54B (23:22Z) and then 'around $3 billion' (10/30 00:52Z); it states it will cover all losses and suspended deposits and withdrawals. Its initial investigation 'may involve a compromised DevOps engineer's laptop' through which an attacker reached a service container and submitted unauthorized withdrawal transactions, with no evidence of a wallet private-key leak. Reported stolen amounts range $5.6B-$6.3B with no confirmed total. This article records the first passive observation of 0x3E2bf1ac6bf0934BBba6456C341c2EcbCED632d5 (named by ZachXBT after the Arbitrum Security Council's ~$30.5M freeze; first tx 10/29 21:29:47Z, 3 EVM txs, 0.0038 ETH) the on-chain verification of the Arbitrum freeze transaction (from = the address itself), and two converging third-party attribution claims (TRM Labs: TraderTraitor; Specter: commingling with the DPRK-attributed Bitget exploit via THORChain).