Glossary Updates16 new terms added to the glossaries · November 12, 2026, 21:23 CET
AI TechDocKnowledge

Glossary · AI engineering and governance

AI bill of materials (AIBOM)

Also known as: AIBOM, AI BOM, ML-BOM, Machine learning bill of materials, Model bill of materials

German: KI-Stückliste (AI Bill of Materials, AIBOM)

In AI governance, an AI bill of materials (AIBOM) is a structured inventory of the AI components behind a product or service: the models with their versions and providers, the training or fine-tuning data sources as far as they are known, prompts or system instructions, inference runtimes and libraries, and the tools an agent is allowed to call. It is modeled on the software bill of materials and is intended to make an AI system's supply chain auditable — who supplied which model, on what data, in which version, and what changed. The practice is emerging: BOM formats such as CycloneDX ML-BOM and the AI and dataset profiles of SPDX 3.0 can carry such data, but as of 2026-10-07 no standard defines the term normatively and no jurisdiction is known to require an AIBOM under that name.

  • AI
  • Industrial AI
  • Technical documentation
  • OT security

In one sentence

An AI bill of materials inventories the models, data sources, prompts and tools in an AI system — an emerging practice modeled on the SBOM.

Example

Before a visual inspection cell goes into production, the system integrator hands over an AIBOM that lists the defect detection model with its version hash, the vendor and license of the pretrained backbone, the image sets used for fine-tuning and the inference runtime, so the plant operator can tell which cells are affected when the model vendor publishes a security advisory.

How it applies

  • Scope first: An AIBOM only helps if its scope is written down — one model, one AI function, or a whole machine. In industrial practice the useful unit is usually the AI function in an asset (inspection, anomaly scoring, setpoint recommendation), because that is what the plant operator validates and releases.
  • Typical entries: model name, provider and license; model version or artifact hash (see Model versioning); base or pretrained model it was derived from; training data and fine-tuning data sources with owner, time window and license; preprocessing and feature code; inference runtime, accelerator libraries and container image; prompts or system instructions for generative components; external tools, APIs and databases an agent may call; deployment location (cloud, edge AI, embedded).
  • Generate it, don't type it: Treat the AIBOM as a build artifact of the MLOps pipeline, exported on every release, not as a one-off spreadsheet. Store it with the release and keep older versions, otherwise you cannot reconstruct which model ran when an incident is investigated.
  • Link it to the asset: In an industrial setting an AIBOM is most useful when it is referenced from the plant asset inventory or carried as a submodel-like attachment to the Asset Administration Shell (AAS) of the asset, so OT teams can answer "which machines run this model version?".
  • Security work: The AIBOM feeds vulnerability management and supply chain security: a CVE in a serialization library or a withdrawn pretrained model becomes a searchable question instead of an archaeology project. It does not replace signing, hardening or access control for model artifacts.
  • Documentation work: Technical writers decide what belongs in customer-facing documentation (model identity, version, intended use, known limitations, data provenance statements) and what stays internal (raw dataset paths, secrets, prompt text that is part of the product's IP). Behavioral changes between model versions belong in the release notes; the AIBOM records that the version changed, not what the change means for the operator.
  • Regulatory context, stated carefully: Regulation (EU) 2024/1689 (AI Act) requires technical documentation for high-risk AI systems along the lines of its Annex IV, and providers of general-purpose AI models have their own documentation duties; neither uses the term "AI bill of materials". The Cyber Resilience Act obliges manufacturers of products with digital elements to draw up a software bill of materials, which covers software dependencies and not AI-specific facts such as training data provenance. Producing an AIBOM is therefore a way to organize evidence — it is not in itself proof that any regulation or standard is met.
  • Status check: Because formats and expectations are still moving, record which format and version you used (for example CycloneDX ML-BOM or SPDX 3.0 with its AI and dataset profiles) and re-check current official guidance before you promise a specific format to a customer or auditor.

AIBOM vs. SBOM

A software bill of materials (SBOM) answers which software components and dependencies are inside a product, with versions and suppliers; it is comparatively mature, has established formats and is named in regulation such as the Cyber Resilience Act. An AIBOM answers the questions an SBOM was never built to answer: which model artifact, trained or fine-tuned on which data, under which license and usage restriction, driven by which prompt, calling which external tools. The two overlap — model serving code, libraries and container images are ordinary software and belong in the SBOM — so the practical pattern is one SBOM for the software supply chain plus AI-specific entries for models, datasets and prompts, not two unrelated documents.

External references

By knowledge.aitechdoc.world · Published October 11, 2026 · Last reviewed

Source: CycloneDX project, Machine Learning Bill of Materials (ML-BOM)

Definitions follow the cited standards and specifications. Where a source is a copyrighted publication, such as an ISO, IEC or EN standard, the definition is a close paraphrase, not a verbatim quotation, so as not to infringe copyright. We recommend reading the original publication. The sections “How it applies” are editorial commentary by AI TechDoc Knowledge and are not part of any standard.

Seen a mistake? Send us a note!