AI glossary Strategy & Business
What is an AI bill of materials (AI BOM)?
AI Bill of Materials (AI BOM)
← All glossary termsA machine-readable inventory of everything an AI system is built from: models, datasets, dependencies, licences and their lineage.
Explained
An AI bill of materials (AI BOM) is an inventory of the components of an AI system: base models and their versions, adapters and fine-tunes, training and evaluation datasets, software libraries, containers, and the licences and provenance of each. It extends the idea of the software bill of materials (SBOM) to the parts that are specific to AI. With it, security, legal and risk teams can see what a system depends on, assess exposure when a model or library has a problem, and answer questions from auditors and regulators. Standards such as OWASP CycloneDX (ML-BOM) and SPDX 3.0 (AI and dataset profiles) define formats for it.
What an AI BOM records
- Models: name, version, supplier, architecture, where they came from.
- Adapters, fine-tunes and prompts that change the model's behaviour.
- Datasets used for training, tuning and evaluation, with their origin and licence.
- Software dependencies: frameworks, libraries, containers, serving runtime.
- Licences and usage restrictions for each component.
- Known limitations, intended use and evaluation results, where available.
Why it matters
AI systems are assembled from parts that change often and come from many suppliers: an open-weight model, a library, a dataset from a third party. When one of them has a vulnerability, a licence issue or a quality problem, you need to know quickly which systems are affected.
An AI BOM also supports compliance. Regulations such as the EU AI Act ask providers of certain systems for technical documentation on data, models and their development. A maintained inventory makes that documentation easier to produce and keep current.
Standards and formats
- OWASP CycloneDX added machine learning BOM (ML-BOM) support in version 1.5 and publishes a guide to AI/ML BOMs.
- SPDX 3.0 includes AI and dataset profiles for describing models and data.
Both are machine-readable, so the BOM can be generated in the build pipeline and checked automatically, instead of being a document written by hand.
Frequently asked questions
What is the difference between an SBOM and an AI BOM?
An SBOM lists the software components of an application. An AI BOM adds what is specific to AI: models, their weights and versions, training data, fine-tunes and their licences and provenance. Many teams generate both from the same pipeline.
Is an AI BOM mandatory?
There is no general legal obligation to produce a document called an AI BOM. However, regulations such as the EU AI Act require technical documentation for certain systems, and many organisations' security and procurement policies ask for SBOMs, so an AI BOM is an efficient way to meet those demands.
What is ML-BOM in CycloneDX?
ML-BOM is CycloneDX's support for describing machine learning components: models, datasets and their configuration, alongside regular software components. It was introduced in CycloneDX 1.5.
Who should own the AI BOM?
Usually the team that builds and operates the AI system produces it, ideally automatically in the build pipeline. Security, legal and risk teams use it, and AI governance defines what it must contain.
Does an AI BOM cover third-party models used through an API?
It should record them, with the provider, model name and version and the terms of use, even if you cannot see their training data. Knowing which external models each system calls is essential when a provider changes or retires a model.
Decision guides that use this term
How to put it to work
Where it shows up at Thinkia
Further reading
Related terms
Sources
- OWASP CycloneDX · Authoritative guide to AI/ML-BOM (PDF)
- OWASP CycloneDX · Introducing CycloneDX v1.5
- Bennet et al. · Implementing AI bill of materials (AI BOM) with SPDX 3.0 (arXiv)
- Building an open AIBOM standard in the wild (arXiv)
Updated: 2026-10-21