FFmpeg vulnerabilities when extending latest AzureML recommended base image (Ubuntu / no Ubuntu Pro)

Francesco Picetti 0 Reputation points Microsoft Employee
2025-12-20T00:17:45.6966667+00:00

I am creating a custom Docker image for Azure Machine Learning by extending one of the most recent recommended base images from the official Azure/AzureML‑Containers repository, pulled from Microsoft Container Registry (MCR).

Starting from this base image, my Dockerfile performs only the following steps:

  1. Pull the latest recommended AzureML base image
  2. Install FFmpeg using the Ubuntu package manager (apt install ffmpeg)
  3. Install Python dependencies using uv (no additional system packages or custom builds)

Despite this minimal and fully documented setup, the resulting image is flagged by container vulnerability scanners due to FFmpeg‑related vulnerabilities originating from Ubuntu Universe packages.

For example: USN-7823-1

From my investigation:

  • The vulnerable FFmpeg packages come directly from the Ubuntu repositories used by the AzureML base image
  • There does not appear to be a patched FFmpeg version available via standard apt updates without enabling Ubuntu Pro / ESM
  • Following AzureML guidance (using recommended base images and standard package installation) still results in images flagged as vulnerable
  • Workarounds such as rebuilding FFmpeg from source or removing features significantly increase image complexity and long‑term maintenance burden

Question: What is the recommended way to address this scenario when building custom AzureML Docker images?

Specifically:

  • Is this a known limitation when using Ubuntu‑based AzureML images?
  • Are users expected to manually rebuild FFmpeg or accept Ubuntu Pro dependencies for compliance‑sensitive environments?
  • Are there plans to refresh AzureML base images with mitigations for these FFmpeg vulnerabilities?
  • Is there official guidance on handling vulnerabilities inherited from Ubuntu Universe packages when no standard patches are available?

Any guidance on the intended or supported approach would be greatly appreciated.

Azure Machine Learning
{count} votes

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.