New ENCFORGE Ransomware Targets AI Model Weights and Training Data via Langflow Flaw

Security researchers at Sysdig have uncovered ENCFORGE, a custom ransomware strain purpose-built to encrypt AI model weights, vector indexes and training data, deployed by an agentic threat actor exploiting a critical Langflow vulnerability. The finding, detailed in a July 20, 2026 report from the Sysdig Threat Research Team, marks what analysts describe as the first known campaign to directly hold an organization's AI assets hostage.