Home / Technology / Ransomware Evolves to Destroy AI Assets
Ransomware Evolves to Destroy AI Assets
27 Jul
Summary
- Ransomware now targets AI models, unlike traditional data.
- Attackers exploit Langflow vulnerability for AI asset destruction.
- Restoring AI models can cost up to $500,000 per model.
A sophisticated ransomware known as ENCFORGE has emerged, specifically targeting and destroying trained AI models. This represents a significant evolution from traditional ransomware, which typically focuses on encrypting data. The attackers exploited a critical vulnerability, CVE-2025-3248, in the Langflow server, enabling them to execute code and deploy the destructive payload.
Sysdig's Threat Research Team documented two distinct campaigns in July 2026. The first involved improvised encryption, while the second deployed ENCFORGE, a Go binary explicitly designed to target AI assets. The ransomware's extension list includes formats like PyTorch, TensorFlow, and GGUF, indicating a deliberate focus on AI model components.
Experts estimate that rebuilding a single production-ready fine-tuned AI model could cost between $75,000 and $500,000. This cost reflects GPU rates and engineering hours, underscoring the substantial business risk. Unlike databases, AI model weights represent learned knowledge that is difficult and expensive to reproduce.
The attackers demonstrated rapid adaptation, building a functional host escape within minutes after their initial attempt failed. This speed of execution compresses defender response times, increasing the blast radius of an attack. The vulnerability used, CVE-2025-3248, has a high CVSS score of 9.8 and was added to CISA's Known Exploited Vulnerabilities catalog in May 2025.
