Pattern: AIBOM

An AI bill of materials emitted at build, recording models, datasets, weights and their provenance in a standard format beside the registry entry.

By Jorge García Aibar
v0.5.0
Updated 2026-09-25
CC BY 4.0

Layer 02 · Inventory & Transparency In the chapter 05 catalogue

One working control on the engineering side of AI governance.

Summary: Generate an AI bill of materials at build for each AI system, recording models, datasets, weights and their provenance in a standard format, and store it with the registry entry. The AIBOM is what the transparency and eval layers read to know what to document and what to test.

In short

AIBOM is an inventory control that generates an AI bill of materials at build for each AI system, recording its models, datasets, weights and their provenance and licences in a standard format. It solves a supply-chain blind spot: the classic SBOM captures software dependencies but not models or data, so without an AIBOM an organisation cannot say which model version, from which provenance, trained on which data, sits inside a given system. Use it when AI systems are assembled from foundation models, fine-tunes, third-party datasets and libraries. The AIBOM is emitted in CycloneDX ML-BOM or the SPDX 3.0 AI profile, attached to the system’s registry entry and regenerated on each build, so it never drifts from the deployed system. The transparency and eval layers read it to know what to document and what to test. Its illustrative mappings are EU AI Act Art. 11 and Art. 53 for GPAI documentation, ISO/IEC 42001, the NIST AI RMF Map function and CSA AICM.

AIBOM at BuildThe build step emits an AI bill of materials that lands on the registry entry and feeds the vulnerability-matching and documentation consumers. If your build does not emit it, you cannot answer what is running. Generated from the Body of Knowledge.Open interactive diagram (opens in a new tab)

Objectives

Make the composition and provenance of an AI system machine-readable, so supply-chain risk and transparency obligations can be answered from an artefact, not reconstructed.

Target users

AI governance engineer, ML engineer, security engineer.

Impacted stakeholders

Model owners, downstream deployers, auditors, procurement.

Relevant principles

Instrument the build to produce its own proof; start from a named failure mode or harm.

Context

AI systems assembled from foundation models, fine-tunes, third-party datasets and libraries, where the classic SBOM captures software dependencies but not models or data.

Problem

Without a bill of materials for models and data, an organisation cannot answer which model version, from which provenance, trained on which data, is inside a given system, so it cannot assess supply-chain risk or produce transparency documentation on demand.

Solution

Emit an AIBOM at build in a standard format, CycloneDX ML-BOM or the SPDX 3.0 AI profile, for example with the OWASP AIBOM generator ¹ (illustrative), covering models, datasets, weights, and their provenance and licences. Attach it to the registry entry and regenerate it on each build so it never drifts from the deployed system.

Consequences

Supply-chain and provenance questions become queries; transparency documents can be generated from the AIBOM. The cost is toolchain integration and keeping provenance metadata accurate.

Agent Registry; Model Card as Control Evidence; Machine-Readable Evidence (OSCAL).

Maps to: EU AI Act Art. 11, Art. 53 (GPAI documentation) · ISO/IEC 42001 · NIST AI RMF (Map) · CSA AICM · Layer 02 Inventory & Transparency.

Function labels follow the NIST AI RMF². Mappings are illustrative, not a claim of conformity.

Sources

[1] “Evolving AI Transparency: the AIBOM generator’s new home at OWASP” (CycloneDX output). OWASP GenAI Security Project. 2025-12-18. https://genai.owasp.org/2025/12/18/evolving-ai-transparency-the-journey-of-the-aibom-generator-and-its-new-home-at-owasp/ (verified: primary)
[2] AI Risk Management Framework (AI RMF 1.0; Govern, Map, Measure, Manage). NIST. 2023-01-26. https://www.nist.gov/itl/ai-risk-management-framework (verified: primary)

Edit this page on GitHub

Cite this pattern

García Aibar, J. (2026). Pattern: AIBOM. In AI Governance Engineering: The Thesis & Body of Knowledge (v0.5.0), chapter 05, Patterns. https://doi.org/10.5281/zenodo.22956197. https://aigovernanceengineer.com/patterns/aibom. CC BY 4.0

BibTeX

@misc{aige2026bok,
  author  = {Jorge García Aibar},
  title   = {{AI Governance Engineering: The Thesis \& Body of Knowledge}},
  chapter = {05. Patterns: AIBOM},
  year    = {2026},
  version = {0.5.0},
  doi     = {10.5281/zenodo.22956197},
  url     = {https://aigovernanceengineer.com/patterns/aibom},
  note    = {Version 0.5.0}
}

What AIBOM connects to

Family

Item

Relation

Related patterns

Agent Registry

Related pattern

Related patterns

Machine-Readable Evidence (OSCAL)

Related pattern

Related patterns

Model Card as Control Evidence

Related pattern

Obligations

CSA AICM v1.1: 247 control objectives across 18 domains

Names this pattern

Obligations

EU AI Act Art. 11: technical documentation (Annex IV)

Names this pattern

Obligations

EU AI Act Art. 53: GPAI provider obligations

Names this pattern

Obligations

EU AI Act Art. 53(1)(c): copyright policy honouring text-and-data-mining reservations

Names this pattern

Obligations

NIST AI RMF MAP

Names this pattern

Controls

AIGE-CTL-ASSURE-012 AI Bill of Materials per Build

Names this pattern

Controls

AIGE-CTL-EVAL-008 Harness and Configuration Attestation

Names this pattern

Controls

AIGE-CTL-DATA-010 Lineage from Training Runs to Admitted Sources

Names this pattern

Controls

AIGE-CTL-AGENT-018 MCP server admission gate

Names this pattern

Controls

AIGE-CTL-ASSURE-010 Model Artefacts Signed at Build and Verified Before Load

Names this pattern

Controls

AIGE-CTL-DATA-009 Signed Snapshot Integrity

Names this pattern

Controls

AIGE-CTL-DATA-003 Training-Data Rights Ledger Row per Source

Names this pattern

Cases

Clearview AI: a face database built by scraping

Names this pattern

Cases

The Garante's ChatGPT order: launch before a lawful basis

Names this pattern

Obligations this pattern evidences

Register rows that list this pattern among the patterns that build their evidence. Mappings are illustrative, not a claim of conformity.

When these obligations apply Each obligation this pattern evidences, on the date it first applies, drawn by its status in the register, and on each later dated step (a square, filled once reached); the dashed line is the date the rows were last reviewed. What comes before the line is already due, so the pattern's evidence is needed now; what comes after it sets the deadline to build it by.

Data table

When the obligations AIBOM evidences apply
Date	Obligation	Step	Status
2025-08-02	EU AI Act Art. 53: GPAI provider obligations	Applies from	In force
2025-08-02	EU AI Act Art. 53(1)(c): copyright policy honouring text-and-data-mining reservations	Applies from	In force
2026-08-02	EU AI Act Art. 53: GPAI provider obligations	Later step: Commission enforcement powers apply	In force
2026-08-02	EU AI Act Art. 53(1)(c): copyright policy honouring text-and-data-mining reservations	Later step: Commission enforcement powers apply	In force
2027-08-02	EU AI Act Art. 53: GPAI provider obligations	Later step: GPAI models placed on the market before 2025-08-02 comply by this date (Art. 111(3))	In force
2027-08-02	EU AI Act Art. 53(1)(c): copyright policy honouring text-and-data-mining reservations	Later step: GPAI models placed on the market before 2025-08-02 comply by this date (Art. 111(3))	In force
2027-12-02	EU AI Act Art. 11: technical documentation (Annex IV)	Applies from	Deferred
2028-08-02	EU AI Act Art. 11: technical documentation (Annex IV)	Later step: Applies to Annex I embedded (product safety-component) systems	Deferred
2030-08-02	EU AI Act Art. 11: technical documentation (Annex IV)	Later step: Deadline for legacy high-risk systems intended for use by public authorities (Art. 111(2))	Deferred
No date	NIST AI RMF MAP		Voluntary
No date	CSA AICM v1.1: 247 control objectives across 18 domains		Voluntary

Obligations this pattern evidences

Cases where it applies

Related controls