JADEPUFFER , a ransomware campaign operated by a large language model, has returned with a new variant called ENCFORGE, specifically designed to destroy AI model artifacts. This campaign utilizes a targeted approach, impacting various file formats critical to machine learning including PyTorch and TensorFlow files. Recovery costs from such attacks could range from $75,000 to $500,000 per model, given the extensive training time involved.
The ransomware exploits a security flaw in the Langflow platform, enabling the operator to evade standard security measures and execute the attack autonomously. Unlike traditional ransomware, JadePuffer's goal is data destruction rather than extortion.