ENCFORGE Ransomware Targets AI Model Files
New threat to AI infrastructure through Langflow RCE attack.
1Emergence of ENCFORGE Ransomware
Sysdig researchers linked a second attack on Langflow servers to the JADEPUFFER operator. This operator is now deploying ENCFORGE, a new ransomware designed to encrypt AI model weights, vector indexes, and training datasets.
2Langflow's Security Vulnerability
Versions of Langflow prior to 1.3.0 expose the /api/v1/validate/code endpoint without authentication, allowing remote attackers to execute arbitrary Python code on the server. This flaw, known as CVE-2025-3248, has a CVSS score of 9.8.
3Targets of ENCFORGE
ENCFORGE ransomware encrypts AI model files such as PyTorch and TensorFlow checkpoints, Hugging Face SafeTensors, ONNX format, and GGUF. The encryption of these files poses a significant threat to AI infrastructure.
4Encryption Mechanism
ENCFORGE uses AES-256-CTR to encrypt file data. A symmetric key is used for each run, wrapped under an RSA-2048 key compiled into the binary. Instead of encrypting entire files, selected regions are encrypted.
5Operator's Adaptation
The JADEPUFFER operator used the Langflow RCE channel to gain access to the host system via the Docker socket. During this process, Python scripts were used to copy and run ENCFORGE ransomware on the host system.
6Impact of Encrypted Files
Reconstructing encrypted AI model files could cost between $75,000 and $500,000 per model due to cloud GPU computing and engineering time. This creates a significant financial risk for AI infrastructure.
7Measures and Conclusions
Updating Langflow and protecting AI model files are critical. ENCFORGE ransomware emerges as a new threat to AI infrastructure, highlighting the need for enhanced security measures to prevent such attacks.
Original article: The Hacker News — ENCFORGE Ransomware AI Model Dosyalarını Hedef Alıyor