Attackers tampered with files served by AOMEI Backupper's own update infrastructure between May and July 2026, modifying what the Windows backup tool would download during windows of just five to ten minutes at a time — so that any machine that happened to check for updates inside one of those narrow intervals would pull down and execute a backdoor.
The incident was reported by The Hacker News on 16 October, with analysis credited jointly to Google Threat Intelligence Group, Mandiant, and ManageEngine's threat analysis team. The findings matter less for the specific malware involved than for the trust model they puncture: the payload arrived through the vendor's own update servers, leaving endpoint protections and user caution — both tuned to flag unsolicited downloads — with little to work with.
Precision, not persistence
What distinguishes this intrusion from typical supply-chain compromises is its brevity. Rather than squatting on an update server for days or weeks — the pattern seen in many software supply-chain incidents, where a persistent implant quietly taints every release — the attackers appear to have swapped the served files for minutes at a time and then reverted them, likely to limit their exposure to detection by the vendor or by security vendors monitoring update traffic. That motive is an inference, not a stated finding in the source report, but the operational effect is documented: five-to-ten-minute tampering bursts, repeated across a May-to-July window.
That restraint has two implications. First, the blast radius is probabilistic: only clients that polled the update server during a tampering window received the malicious files, so the affected population is a subset of AOMEI Backupper users rather than the entire install base — but it is also difficult to enumerate after the fact, since no poisoned version number or poisoned release exists to trace back through patch histories. Defenders must reason about exposure windows, not version inventories. Second, the technique leans on the assumption that update channels are inherently trustworthy — an assumption embedded in both user behaviour and many allowlisting configurations.
The risk profile is amplified by what AOMEI Backupper is. Backup agents typically run with elevated privileges and hold access to recovery data — precisely the resources an incident responder leans on when containing a compromise. A backdoor delivered inside a backup tool therefore undermines the recovery path itself, potentially giving an attacker visibility into, or influence over, the copies an organisation would rely on to restore clean systems. Such an implant could survive a routine remediation effort and contaminate the very images used to rebuild clean machines. Defenders cannot treat "it came from the vendor" as evidence that a binary is benign.
The source report does not identify which AOMEI versions or builds were affected, and contains no remediation announcement from the vendor. Organisations running the product should check AOMEI's official channels directly for current guidance rather than relying on any single account of the incident.
What HK IT teams should check now
For administrators managing Windows fleets in Hong Kong, the practical takeaway is a short integrity audit of the backup toolchain:
- Pin and verify signatures. Confirm that update binaries carry valid Authenticode signatures issued to AOMEI, and configure endpoints or proxy appliances to block execution of update payloads that fail signature validation — not merely to warn. Note that whether signature verification alone would have defeated this payload depends on whether the signing key was compromised or only the served files were swapped; the source report does not say.
- Baseline expected behaviour. Record which domains and servers your backup agents contact for updates, and alert on any deviation, including certificate changes on those endpoints.
- Log update activity. Maintain records of when each endpoint last checked for and applied updates, so that any machine falling inside a suspected tampering window can be identified and isolated for inspection.
- Scan for persistence, not just payload. Treat backup agents as high-value: review their scheduled tasks, services, and loaded modules for anomalies, and validate that recovery media and restore paths still produce clean systems.
- Compensate with isolation. Where possible, segment backup infrastructure from general workstation networks so that a compromised agent cannot reach unrelated recovery assets.
The core lesson of the AOMEI incident is straightforward: an official update channel is a distribution mechanism, not a guarantee. In supply-chain defence, provenance is a claim to be verified, not a property to be assumed. ```
```yaml
title: "AOMEI Backupper 更新伺服器遭竄改 短暫時段內投放後門" date: 2026-10-16T19:05:00 author: HKLUG Team source_url: https://thehackernews.com/2026/10/attackers-tampered-with-aomei.html tags: [supply-chain, malware, backup, windows, aomei]
攻擊者於 2026 年 5 至 7 月期間,竄改 AOMEI Backupper 自有更新基礎設施所發放的檔案,改變這款 Windows 備份工具在僅五至十分鐘的短暫時段內下載的內容 —— 任何恰在這些時段內檢查更新的電腦,便會下載並執行一個後門程式。
事件由 The Hacker News 於 10 月 16 日報道,分析由 Google Threat Intelligence Group、Mandiant 與 ManageEngine 的威脅分析團隊共同負責。事件的意義,與其說在於所涉的惡意程式,不如說在於它戳破的信任模型:payload 經由供應商自己的更新伺服器送達,令端點防護與用户警惕 —— 兩者本來都是針對未經邀請的下載而設 —— 幾無用武之地。
精準,而非長期潛伏
這次入侵與典型供應鏈攻擊的分別,在於它的短暫。攻擊者並非在更新伺服器上駐守數天至數星期 —— 那是許多軟件供應鏈事故的典型模式,持久性植入程式悄然污染每一次發佈 —— 而似乎只是把伺服器上的檔案換掉幾分鐘,隨後還原,很可能是為了減少被廠商或監察更新流量的保安廠商偵測到的機會。這個動機屬推斷,並非來源報道明載的結論;但實際效果則有清楚記錄:為時五至十分鐘的竄改突發,在 5 至 7 月間反覆出現。
這樣的節制有兩重含意。首先,影響範圍屬機率性質:只有在竄改時窗期間向更新伺服器查詢的用戶端,才會收到惡意檔案,因此受影響的只是 AOMEI Backupper 用戶的一部分,而非整個安裝基數 —— 但事後亦難以清點,因為根本沒有遭污染的版本號或發佈可供追查修補紀錄。防禦者必須按暴露時窗思考,而非按版本清單。其次,這套手法倚靠的,是「更新渠道天然可信」這一假設 —— 而該假設已嵌入用户行為與不少 allowlist 設定之中。
風險之所以加劇,在於 AOMEI Backupper 本身的性質。備份代理程式通常以提升權限執行,並可存取復原資料 —— 正是事故應變人員控制事故時所倚靠的資源。因此,在備份工具內投放後門,等同削弱復原路徑本身,或令攻擊者得以觀看、甚至影響組織用來還原乾淨系統的副本。這類植入程式可能在例行補救行動中存活,並污染用以重建乾淨電腦的映像。防禦者不能把「來自供應商」當成二進位檔屬良性的證據。
來源報道沒有指出哪些 AOMEI 版本或建置版本受影響,亦沒有載有廠商的補救公告。使用該產品的機構應直接查閱 AOMEI 官方渠道的最新指引,而非依賴任何單一報道。
香港 IT 團隊現階段應檢查的事項
對管理香港 Windows fleet 的管理員而言,實務上應對備份工具鏈作一次簡短的完整性審視:
- 固定並核對簽章。 確認更新用的二進位檔帶有 AOMEI 簽發的有效 Authenticode 簽章,並設定端點或代理伺服器,攔截執行未通過簽章驗證的更新 payload —— 而不只是發出警告。要注意,單靠簽章驗證能否擋下這次 payload,取決於簽章金鑰本身有否被入侵,抑或只是伺服器上的檔案被掉包;來源報道沒有說明。
- 為預期行為建立基準。 記錄備份代理程式更新時會連繫的網域名稱與伺服器,並在出現任何偏離時告警,包括該等端點的憑證變更。
- 記錄更新活動。 保存每部端點最後一次檢查及套用更新的時間紀錄,以便識別任何可能落入疑似竄改時窗的電腦,並將其隔離作檢查。
- 搜尋持久化,而非只看 payload。 把備份代理程式視為高價值目標:檢視其排程工作、服務及載入模組有否異常,並驗證復原媒體與還原路徑仍能產生乾淨系統。
- 以隔離作補償。 可行時,把備份基建與一般工作站網絡分段,令被入侵的代理程式無法接觸其他復原資源。
AOMEI 事件的核心教訓很直接:官方更新渠道是發放機制,不是保證。在供應鏈防禦中,provenance 是一項要核實的聲稱,而非可以假定的性質。
攻擊者於 2026 年 5 至 7 月期間,竄改 AOMEI Backupper 自有更新基礎設施所發放的檔案,改變這款 Windows 備份工具在僅五至十分鐘的短暫時段內下載的內容——任何恰在這些時段內檢查更新的電腦,便會下載並執行一個後門程式。
事件由 The Hacker News 於 10 月 16 日報道,分析由 Google Threat Intelligence Group、Mandiant 與 ManageEngine 的威脅分析團隊共同負責。事件的意義,與其說在於所涉的惡意程式,不如說在於它戳破的信任模型:payload 經由供應商自己的更新伺服器送達,令端點防護與用戶警惕——兩者本來都是針對未經邀請的下載而設——幾無用武之地。
精準,而非長期潛伏
這次入侵與典型供應鏈攻擊的分別,在於它的短暫。攻擊者並非在更新伺服器上駐守數天至數星期——那是許多軟件供應鏈事故的典型模式,持久性植入程式悄然污染每一次發佈——而似乎只是把伺服器上的檔案換掉幾分鐘,隨後還原,很可能是為了減少被廠商或監察更新流量的保安廠商偵測到的機會。這個動機屬推斷,並非來源報道明載的結論;但實際效果則有清楚記錄:為時五至十分鐘的竄改突擊,在 5 至 7 月間反覆出現。
這樣的節制有兩重含意。首先,影響範圍屬機率性質:只有在竄改時窗期間向更新伺服器查詢的用戶端,才會收到惡意檔案,因此受影響的只是 AOMEI Backupper 用戶的一部分,而非整個安裝基數——但事後亦難以清點,因為根本沒有遭污染的版本號或發佈可供追查修補紀錄。防禦者必須按暴露時窗思考,而非按版本清單。其次,這套手法倚靠的,是「更新渠道天然可信」這一假設——而該假設已嵌入用戶行為與不少 allowlist 設定之中。
風險之所以加劇,在於 AOMEI Backupper 本身的性質。備份代理程式通常以提升權限執行,並可存取復原資料——正是事故應變人員控制事故時所倚靠的資源。因此,在備份工具內投放後門,等同削弱復原路徑本身,或令攻擊者得以觀看、甚至影響組織用來還原乾淨系統的副本。這類植入程式可能在例行補救行動中存活,並污染用以重建乾淨電腦的映像。防禦者不能把「來自供應商」當成二進位檔屬良性的證據。
來源報道沒有指出哪些 AOMEI 版本或建置版本受影響,亦沒有載有廠商的補救公告。使用該產品的機構應直接查閱 AOMEI 官方渠道的最新指引,而非依賴任何單一報道。
香港 IT 團隊現階段應檢查的事項
對管理香港 Windows 機群的管理員而言,實務上應對備份工具鏈作一次簡短的完整性審視:
- 固定並核對簽章。 確認更新用的二進位檔帶有 AOMEI 簽發的有效 Authenticode 簽章,並設定端點或代理伺服器,攔截執行未通過簽章驗證的更新 payload——而不只是發出警告。要注意,單靠簽章驗證能否擋下這次 payload,取決於簽章金鑰本身有否被入侵,抑或只是伺服器上的檔案被掉包;來源報道沒有說明。
- 為預期行為建立基準。 記錄備份代理程式更新時會連繫的網域名稱與伺服器,並在出現任何偏離時告警,包括該等端點的憑證變更。
- 記錄更新活動。 保存每部端點最後一次檢查及套用更新的時間紀錄,以便識別任何可能落入疑似竄改時窗的電腦,並將其隔離作檢查。
- 搜尋持久化,而非只看 payload。 把備份代理程式視為高價值目標:檢視其排程工作、服務及載入模組有否異常,並驗證復原媒體與還原路徑仍能產生乾淨系統。
- 以隔離作補償。 可行時,把備份基建與一般工作站網絡分段,令被入侵的代理程式無法接觸其他復原資源。
AOMEI 事件的核心教訓很直接:官方更新渠道是發放機制,不是保證。在供應鏈防禦中,provenance 是一項要核實的聲稱,而非可以假定的性質。
