Hackers Could Weaponize GGUF Models to Achieve RCE on SGLang Inference Servers

A critical SSTI vulnerability (CVE-2026-5760) in the SGLang framework allows attackers to achieve remote code execution by embedding malicious templates in GGUF model metadata; when a compromised model is loaded and the /v1/rerank endpoint renders the template using an insecure Jinja2.Environment, arbitrary OS commands can be executed — a PoC exists and administrators are advised to avoid untrusted models and adopt sandboxed template rendering.