Arbitrary Code Execution in BentoML Python Library for AI Applications

CVE-2026-44345

What is CVE-2026-44345?

BentoML, a Python library designed for building optimized online serving systems for AI applications, contains a vulnerability that allows attackers to exploit improper handling of Dockerfile templates. Specifically, prior to version 1.4.39, the affected Dockerfile template does not escape or validate multi-line inputs from a bento.yaml file. This flaw can lead to the injection of arbitrary Dockerfile directives into the generated Dockerfile, enabling malicious users to execute unauthorized commands on the host system during the containerization process. Users are advised to update to version 1.4.39 or later to mitigate this risk.

References