New ENCFORGE Ransomware Targets AI Model Files in Langflow RCE Attack

Researchers linked the JADEPUFFER operator to a campaign exploiting an unauthenticated Langflow RCE (CVE-2025-3248) to deploy ENCFORGE, a UPX-packed Go ransomware that selectively encrypts AI model checkpoints, vector indexes, and training data. The analysis details the binary (packed/unpacked hashes, embedded RSA key, YARA), the attack chain (credential harvesting, Docker socket host breakout, privileged container and nsenter to run the binary), observed indicators (sha256, protonmail contact), and recommended mitigations including patching Langflow, rotating credentials, removing or restricting docker.sock, and protecting model artifacts.