Microsoft has detailed a novel automated cloud attack technique that compresses the timeline for catastrophic data loss to roughly 120 seconds—a speed that renders many existing incident response playbooks structurally ineffective. The research, published by Microsoft and corroborated by findings from Sysdig, centers on the Storm-3168 threat actor, linked to the JADEPUFFER group, which exploits compromised service principal credentials to permanently delete Azure Blob Storage containers with alarming efficiency.
The technique represents what Sysdig, which first observed the activity in July 2026, has termed "agentic ransomware"—a class of semi-autonomous tooling that executes complex cloud attacks without prolonged manual operator involvement. Rather than encrypting data for extortion, this model targets outright destruction through the identity and access management layer, a shift that fundamentally changes the risk calculus for cloud-dependent organizations.
At the heart of the attack surface are service principals: non-human identities used by applications, CI/CD pipelines, and automation to access cloud resources. These identities routinely accumulate broad, persistent permissions and are frequently exempt from multi-factor authentication—making them, in Microsoft's characterization, "skeleton keys" that grant traversing and destructive power across a cloud management plane.
Once adversaries obtain service principal credentials, the operational tempo becomes devastating. Microsoft's forensic analysis confirms Storm-3168 can progress from initial access to irreversible deletion of Blob Storage containers in approximately two minutes. At that velocity, the window for detection-and-intervention-based response—long the backbone of enterprise security posture—effectively closes before defenders can act.
For any organization running Azure-hosted workloads, the implication is clear: the attack surface here is not a software vulnerability but a governance failure around machine identities. Microsoft's advisory outlines four essential preventative measures:
- Enforce Least-Privilege Access: Continuously audit and restrict service principal permissions to the absolute minimum required for their function.
- Apply Conditional Access Policies: Where supported, bind service principal operations to trusted networks, locations, or conditions to limit abuse potential.
- Implement Continuous Monitoring: Deploy tools such as Microsoft Defender for Cloud to flag anomalous service principal behavior, including mass deletion attempts or unusual API call patterns.
- Practice Rigorous Credential Hygiene: Treat service principal secrets with the same rigor as production database credentials—rotate them frequently, store them securely, and never hard-code them in source repositories or configuration files.
Organizations should also evaluate migrating long-lived service principal secrets to more secure alternatives where feasible, such as Azure managed identities or workload identity federation, both of which reduce reliance on manually managed credentials.
The emergence of agentic ransomware marks a structural shift in cloud threat operations. Non-human identities—long overlooked in favor of user-account-centric security programs—now represent a primary and demonstrably exploitable attack vector. Proactive identity governance is no longer a best-practice recommendation; for any organization with meaningful cloud dependence, it is the first and most critical line of defense.
Microsoft近日詳細描述了一種新穎的自動化雲端攻擊技術，將災難性數據損失的時間線壓縮至約120秒——這一速度使許多現有的事件回應劇本在結構上形同失效。這項由Microsoft發表並經Sysdig研究結果驗證的研究，聚焦於與JADEPUFFER組織關聯的威脅行為者Storm-3168，該行為者利用被洩露的服務主體（service principal）憑證，以驚人的效率永久刪除Azure Blob Storage容器。
此技術代表了Sysdig（於2026年7月首次偵測到該活動）所稱的「自主式勒索軟件」（agentic ransomware）——一種無需操作人員長時間介入即可執行複雜雲端攻擊的半自動化工具類別。與其透過加密數據進行勒索，此模式直接針對身份與存取管理（IAM）層面進行徹底破壞，這一轉變從根本上改變了雲端依賴型組織的風險評估。
攻擊面的核心是服務主體（service principals）：由應用程式、CI/CD管道及自動化工具用以存取雲端資源的非人類身份。這些身份通常積累廣泛且持久的權限，且經常獲豁免於多因素驗證——使其成為Microsoft所形容的「萬能鑰匙」（skeleton keys），能夠橫跨雲端管理平面賦予遍歷及破壞能力。
一旦攻擊者取得服務主體憑證，其操作節奏便變得極具毀滅性。Microsoft的取證分析證實，Storm-3168能在大約兩分鐘內從初始存取進展至Blob Storage容器的不可逆刪除。在此速度下，基於偵測與介入的回應窗口——長期以來作為企業安全態勢的支柱——在防禦者採取行動前便已實質關閉。
對任何運行Azure託管工作負載的組織而言，其啟示顯而易見：此處的攻擊面並非軟件漏洞，而是圍繞機器身份的治理失敗。Microsoft的通告概述了四項必要的預防措施：
- 強制最小權限存取： 持續審計並限制服務主體權限至其功能所需的絕對最低限度。
- 應用條件式存取策略： 在支援的情況下，將服務主體操作綁定至受信任的網絡、位置或條件，以限制濫用的可能性。
- 實施持續監控： 部署如Microsoft Defender for Cloud等工具，以標記異常的服務主體行為，包括大量刪除嘗試或異常API調用模式。
- 嚴格實踐憑證衛生： 以同等嚴謹度對待服務主體密鑰與生產數據庫憑證——定期輪替、安全存儲，且絕不硬編碼於原始碼儲存庫或設定檔中。
組織亦應評估在可行情況下，將長效服務主體密鑰遷移至更安全的替代方案，例如Azure託管身份（managed identities）或工作負載身份聯合（workload identity federation），兩者皆可減少對手動管理憑證的依賴。
自主式勒索軟件的出現標誌著雲端威脅操作的結構性轉變。非人類身份——長期以來因以用戶帳戶為中心的安全計劃而被忽視——現已成為主要且可被明確利用的攻擊向量。主動式身份治理不再是最佳實踐建議；對任何具備顯著雲端依賴性的組織而言，它是第一道也是最關鍵的防線。