A new strain of ransomware has been identified that exclusively targets the core intellectual property of AI operations: model weights, training datasets, and vector indexes.
Security researchers at Sysdig have uncovered the attack, linking it to the JADEPUFFER threat actor previously associated with AI-agent-driven campaigns. The group has now deployed a specialized tool, ENCFORGE, a compiled Go ransomware built to hunt for and encrypt critical AI asset files across a compromised host's filesystem.
This marks a significant evolution in ransomware tactics. Rather than targeting conventional documents or databases, ENCFORGE seeks out specific file types like PyTorch model weights (.pt, .pth), binary data files (.bin), and vector databases. The objective is clear: to directly cripple an organization's ability to develop and deploy artificial intelligence.
The attack vector in this case was a known remote code execution (RCE) flaw in Langflow, an open-source framework for building AI workflows. The incident underscores the expanding attack surface within the AI tooling ecosystem, where platforms optimized for speed and experimentation may run with insecure default settings.
The operational impact can be profound. Encrypting model weights and training data effectively holds AI functions hostage. Recovering these assets is not as simple as restoring from a backup; it often demands the costly and time-consuming process of retraining models from scratch. This transforms a data loss event into a direct hit on innovation capacity and competitive advantage.
In light of this threat, a hardened defensive posture is required. Immediate actions include patching the specific Langflow vulnerability and deploying monitoring configured to detect encryption activity targeting AI-specific file extensions. Fundamentally, organizations must develop backup strategies that explicitly archive AI intellectual property, storing these backups in an immutable, isolated environment to guarantee recoverability.
The emergence of ENCFORGE confirms that adversaries are adapting their tools to strike where it hurts most. For enterprises heavily invested in AI, securing the development and training pipeline is no longer optional—it is a critical pillar of operational resilience.
一種新型勒索軟件已被識別，其專門針對AI運營的核心知識產權：模型權重、訓練數據集和向量索引。
Sysdig的安全研究人員揭露了這起攻擊，並將其與先前與AI代理驅動活動相關的威脅行為者JADEPUFFER聯繫起來。該組織現在部署了一款專門工具ENCFORGE，這是一個用Go語言編譯的勒索軟件，旨在在受感染主機的文件系統中搜尋並加密關鍵AI資產文件。
這標誌著勒索軟件策略的重大演變。ENCFORGE並非針對常規文檔或數據庫，而是搜尋特定文件類型，如PyTorch模型權重（.pt、.pth）、二進位數據文件（.bin）和向量數據庫。其目標明確：直接癱瘓組織開發和部署人工智能的能力。
此次攻擊向量是Langflow中一個已知的遠端代碼執行（RCE）漏洞，Langflow是一個用於構建AI工作流程的開源框架。此事件凸顯了AI工具生態系統中不斷擴大的攻擊面，那些優化於速度和實驗的平台可能以不安全的預設設置運行。
運營影響可能非常深遠。加密模型權重和訓練數據實際上是將AI功能扣為人質。恢復這些資產並不像從備份中還原那麼簡單；它通常需要耗時且昂貴的過程，從頭開始重新訓練模型。這將數據丟失事件轉變為對創新能力和競爭優勢的直接打擊。
鑒於此威脅，需要強化的防禦姿態。立即行動包括修補特定的Langflow漏洞，並部署監控以偵測針對AI特定文件擴展名的加密活動。從根本上講，組織必須制定明確歸檔AI知識產權的備份策略，並將這些備份存儲在不可變、隔離的環境中以確保可恢復性。
ENCFORGE的出現證實了對手正在調整其工具以打擊最痛處。對於重金投入AI的企業而言，保護開發和培訓管道不再是可選的——它是運營韌性的關鍵支柱。