🚀 Built for the community. Built for AI transparency.
We announced something pretty exciting during RSAC 2025: the release of the first-ever open-source tool to generate AI SBOMs for Hugging Face models. This tool, created by me (Helen Oakley) and Dmitry Raidman—two folks who’ve been knee-deep in application security, software supply chain, and AI for quite some time—marks a major step forward in bringing transparency and security into the world of Artificial Intelligence.
The reception from the community? Overwhelmingly positive. It’s been incredibly energizing to see this project spark real conversations about accountability in AI. And honestly, this is just the beginning.
So, What’s an AI SBOM and Why Should You Care?
Let’s rewind for a second. Before we talk AI SBOM, let’s talk SBOM (Software Bill of Materials). Think of it like an ingredient list on a box of cookies—or a drink label (shoutout to Dmitry’s beerBOM analogy from RSAC). If you care about what goes into your food or drink, you should care just as much about what goes into your software.
An SBOM is that ingredient list, but for code. It helps organizations identify what software components they’re using, so they can find and fix vulnerabilities faster, manage licensing, and stay compliant. Two major formats are widely adopted today: CycloneDX and SPDX.
Now, AI SBOM takes this a step further. It includes not just software dependencies, but also information unique to AI systems:
- Where the model came from
- What architecture and frameworks it uses
- How and with what data it was trained
- Energy consumption and licensing
That visibility is critical. AI SBOMs help with transparency, evidence collection, risk management, and secure-by-design development. They’re also becoming essential for aligning with emerging standards and regulations like the EU AI Act, ISO 42001, and other.
Why Hugging Face?
If you're building with AI, chances are you've used a model from Hugging Face. It's become the go-to platform for sharing, discovering, and deploying machine learning models—especially in natural language processing and generative AI.
📊 Here are some highlights:
- Over 900,000 models, 200,000 datasets, and 300,000 demo apps (Spaces) are hosted on the Hugging Face Hub.
- More than 50,000 organizations utilize Hugging Face, including major tech companies like Meta, Google, Microsoft, and Amazon.
- Popular models like BERT and DistilBERT receive over 100,000 weekly downloads.
- The Transformers library by Hugging Face has garnered over 71,800 stars on GitHub, reflecting its popularity among developers.
By focusing our tool on Hugging Face models, we're addressing a significant portion of the AI landscape. Given Hugging Face's widespread adoption, enhancing transparency and security here tackles a substantial chunk of industry risk.
How the AI SBOM Generator Helps
Our tool was built to close that gap.
It generates machine-readable AI SBOMs from Hugging Face model metadata in CycloneDX format, directly supporting the real-world use cases defined by the AI SBOM Tiger Team (part of the CISA.gov SBOM initiative).
It effectively connects the dots between technical standards (CycloneDX/SPDX), community-driven use cases, and the operational needs of organizations that want to manage AI responsibly.
One unique bonus use case? This tool also helps with creating consistent model cards—standardized documentation for models that can help internal teams and external consumers understand what’s inside and how to evaluate trust.
How It Works (No Deep Tech Knowledge Required)
We designed the AI SBOM Generator to be simple, visual, and accessible:
- Just enter a Hugging Face model ID or URL
- The tool pulls available metadata and displays it in a clean, human-friendly view
- You can download the actual AI SBOM .json file in CycloneDX format
- It also calculates a completeness score based on how much metadata is available for selected model
About that score: it ranges from 0 to 100 and evaluates how well the model card supports transparency across key SBOM fields (like model name, version, license, training info, etc.). As the industry continues to evolve, and more AI SBOM fields get standardized, the scoring logic will grow with it—ensuring it stays relevant and useful. Further technical documentation can be found on GitHub.
Open Source, Open Collaboration: What’s Next
This tool is open-source for a reason: we want this to be a community-driven effort. We’re inviting feedback, ideas, and collaboration from across the AI, security, and software engineering worlds.
As the AI SBOM Tiger Team continues to define and refine guidance around use cases, we’ll continue evolving the tool to match. That includes enhancing the scoring logic, expanding automation features, and yes—eventually adding support for SPDX and AI SBOM enrichment capabilities.
We’re not making wild promises—but the vision is clear: help shape a future where every AI component is as trackable and trustworthy as the code it runs on.
Try the Tool Today—And Let Us Know What You Think!
👉 http://owasp-genai-aibom.org/
🔧 Also listed in the CycloneDX Tool Center: https://cyclonedx.org/tool-center
Transparency in AI isn’t optional—it’s the foundation for securing our shared future.