What is an AI-BOM?
An AI Bill of Materials is a structured inventory of every model, dataset, tool, prompt asset, and dependency used by a production AI system. Auditors ask for it; AI-SPM produces it as a continuous artefact.
Last reviewed September 2026
The gap AI Bill of Materials closes
AI systems compose more dependencies than traditional applications: foundation models with versions and providers, fine-tuned variants, embeddings, vector stores, RAG corpora, tools and MCP servers, prompt templates, evaluation datasets. Reconstructing this composition for an audit at incident time is too late.
How Penaxtra delivers AI Bill of Materials
Treat the AI-BOM as a first-class artefact, assembled from the same inventory that drives runtime policy rather than rebuilt by hand for each audit. Penaxtra exports it as a versioned JSON document or a flat CSV, from the console or the API, with provenance on every component.
AI Bill of Materials capabilities
Datasets: RAG corpora (source registry + last-indexed timestamp + sensitivity), evaluation sets, training/fine-tune datasets where customer-owned
Tools and MCP servers: tool ID, scope, transport, owning team, last-reviewed date
Prompt assets: system prompt registry (template ID + version + last-modified), prompt gateways
Dependencies: vector store engine and version, embedding model and version, retrieval pipeline components
Provenance: who declared each entry, when it was last verified, what scan run produced it
AI Bill of Materials compliance mapping
EU AI Act Article 11 (technical documentation, Annex IV), ISO/IEC 42001 A.6.2 (resources), NIST AI 600-1 (GOVERN 1.4 inventory, MAP 1.5 third-party). NIST SP 800-218A maps the AI-BOM concept onto the existing SBOM practice (SSDF PO.5).
Frequently asked
How does an AI-BOM differ from a traditional SBOM?
An SBOM lists software components and licenses. An AI-BOM extends that to model artefacts (foundation, fine-tune, weights), training and retrieval datasets, prompt assets, tools, and runtime dependencies that influence model output. The schemas complement each other.
Is there a public AI-BOM standard?
No single standard yet. NIST AI 600-1 and ISO/IEC 42001 require an inventory but do not prescribe a schema. CycloneDX (industry consortium) has a draft AI/ML extension worth tracking. The AI-BOM exported by Penaxtra carries its own schema name and version field, so a consumer can pin the shape it parses; see the AI-BOM API reference.
Who owns the AI-BOM internally?
Typically the AI platform team produces it, the security team reviews it, and the GRC team submits it to auditors. AI-SPM generates the artefact continuously and assigns review owners per asset class.