Hi everyone. I'm new to the whole cloud security space, but I've been reading a lot about supply chain risks, especially with AI/ML projects.
I'm starting to experiment with some open-source models in containers, and I'm worried about all the dependencies (Python packages, base images, etc.). I've seen Wiz mentioned for cloud security, but can it actually scan things like a Hugging Face model's dependency tree or a custom Dockerfile for an inference service?
Would love to hear if anyone is using it for this specific use case. What does it catch, and what might it miss? Any insights would be really appreciated
I'm also exploring this space. From what I've seen, Wiz can scan container images for known vulnerabilities in the OS packages, which would cover your base image. But I'm not sure it goes deep into the Python dependency tree from a requirements.txt or a Pipfile.lock for custom code.
For Hugging Face models specifically, you might need something that can parse their model cards and configs. Have you looked at any dedicated SBOM tools for ML, like maybe those that generate a software bill of materials for the pipeline? I'd be curious if Wiz integrates with those outputs.
You're right about the OS packages, but Wiz actually does go deeper than that now. It can parse requirements.txt and Pipfile.lock if you're using their container image scanning. The catch is it needs those files to be present in the image at scan time, which they often aren't in final production layers. So you have to scan your build-stage image, not just the slim runtime one.
On the SBOM integration point, they do have an API for ingesting external SBOMs in CycloneDX or SPDX format. So you could generate an SBOM for your ML pipeline with a dedicated tool (like Syft or something ML-specific) and then feed it in. But that's an extra step, and you're now managing another toolchain. It works, but it's not as seamless as having it built-in.
Automate everything. Twice.