Models
Which models are used, where they came from, their versions and how they were fine-tuned.
An AI bill of materials is the ingredient list for an AI system. It is the foundation for securing, governing and proving the AI you run.
An AI‑BOM, or AI bill of materials, is a structured inventory of everything that makes up an AI system: its models, the data used to train and run them, the code and frameworks around them, the infrastructure they depend on and the processes that produce them.
Where a software bill of materials (SBOM) lists code components, an AI‑BOM captures the AI-specific parts that SBOM tools cannot detect.
clusters of minimum elements in the G7 cybersecurity agencies’ 2026 guidance on SBOMs for AI, from models and datasets to infrastructure
malicious AI models identified on Hugging Face in JFrog’s 2026 software supply chain report
Source: JFrog, 2026
average share of hallucinated packages in code from open-source models (at least 5.2% for commercial models), across 576,000 samples
Think of it as an exploded view of an AI system: every layer, recorded and traceable.
Which models are used, where they came from, their versions and how they were fine-tuned.
The datasets behind training, fine-tuning, evaluation and retrieval, with their sources.
The AI libraries, SDKs and agent frameworks the system relies on.
Where the system runs: cloud services, endpoints and hosting.
How the system is built and changed, and who is accountable for it.
Known vulnerabilities and assessment results linked to the components they affect.
They are complementary. Most organizations need both, because each answers a different question.
| SBOM | AI‑BOM | |
|---|---|---|
| Focus | Software components and packages | AI models and their data, infrastructure and processes |
| Answers | What code is in this application? | What AI is in this system, and what is it made of? |
| Detects AI components | No | Yes |
| Supports | Software supply chain security | AI supply chain risk, AI governance and AI compliance |
Four kinds of tools touch AI‑BOMs. Standards define the format; the other three produce documents in it, with very different coverage.
| Tool category | What it is | Ask before you rely on it |
|---|---|---|
| Open standards and formats | CycloneDX, an OWASP standard published by Ecma International as ECMA-424, has ML‑BOM support for models, datasets and their dependencies. SPDX 3.0, released in April 2024, added AI and Dataset profiles. Neither scans anything: they define the document. | Which format do your customers, auditors and tools already accept? Many teams need both. |
| Open-source generators | Command-line tools and libraries that write a CycloneDX or SPDX document for one project or model, usually run by hand or in a build. | Who runs it on every repository, keeps it current and owns what it finds? |
| SBOM and software composition analysis tools | The SBOM tooling already in your pipeline. Some are adding detection of AI libraries and model files. | Does it record the models, datasets and system prompts themselves, or only the packages that load them? |
| AI security and governance platforms | Platforms that generate AI‑BOMs continuously across the estate and connect each one to an owner, risk findings and compliance evidence. Cranium AI is in this category. | Does it cover agents and MCP servers, and does it export CycloneDX and SPDX so the data is never locked in? |
An AI‑BOM turns shadow AI into a known, owned inventory, the first step of the AI Trust Loop.
See DiscoverWhen a model or library is found to be unsafe, an AI‑BOM shows exactly which systems use it.
Frameworks such as the EU AI Act, NIST AI RMF and ISO 42001 expect you to know what your AI is built from.
Vendors can attach AI‑BOMs to Cranium AI Cards, so customers see what is inside the AI they buy.
See AI CardsDetect AI, CodeSensor, CloudSensor and AgentSensor find AI wherever it lives.
Cranium ONE builds AI‑BOMs from what it finds, one per repository.
Each AI‑BOM has an accountable owner, so every component has someone answerable for it.
CodeSensor does not scan infrastructure, so infrastructure components are added through self-attestation.
Attach AI‑BOMs to AI Cards to prove posture to auditors, customers and partners.
Straight answers, kept current. Last updated .
Ask us anything →An AI‑BOM, or AI bill of materials, is a structured inventory of an AI system’s models, data, code, infrastructure and processes.
An SBOM lists software components. An AI‑BOM covers AI models and their data, infrastructure and processes. SBOM tools cannot detect AI components, so an SBOM alone leaves AI out.
AI bill of materials. You will also see it written as AIBOM or AI BOM.
In Cranium ONE, AI‑BOMs are generated per repository, each with an assigned owner, so larger organizations typically have many.
Yes. Cranium ONE generates AI‑BOMs automatically from what Detect AI, CodeSensor, CloudSensor and AgentSensor discover. Infrastructure components that are not scanned are added through self-attestation.
Four kinds of tools are involved. Open standards, CycloneDX ML‑BOM and the SPDX 3.0 AI and Dataset profiles, define the format. Open-source generators and SBOM tools produce documents for individual projects. AI security and governance platforms such as Cranium ONE generate AI‑BOMs continuously from your repositories and export them as CycloneDX or SPDX.
Security teams use it to find exposure, governance and compliance teams use it as evidence, and vendor-risk teams request it from suppliers through AI Cards.
Primary sources for the figures and frameworks cited on this page. Last verified .
See how Cranium ONE discovers your AI and builds AI‑BOMs automatically.
Cookie Policy · Privacy Policy
Global Privacy Control detected. Opt-out request honored: marketing tags stay off in this browser.