What happened
CISA, together with G7 partners from Canada, France, Germany, Italy, Japan, the United Kingdom, and the European Union, published joint guidance titled 'Software Bill of Materials for AI – Minimum Elements' on June 16, 2026. The document defines seven information 'clusters' for an AI SBOM: Metadata, System Level Properties, Models, Datasets Properties, Infrastructure, Security Properties, and Key Performance Indicators. It extends the existing SBOM framework (rooted in EO 14028/NTIA 2021) to cover AI-specific components: model lineage, training dataset provenance, adversarial robustness controls, and prompt-injection risk mitigations. The guidance is voluntary and does not create new legal requirements, but explicitly maps onto EU AI Act Articles 11 and 13/Annex IV technical documentation obligations.
Why it matters
This is the first G7-consensus definition of what an AI SBOM must contain. It establishes a de-facto international baseline for AI supply-chain transparency that is already shaping vendor contracting, procurement questionnaires, and incident-response checklists. Organisations that produce or procure AI systems — especially those selling to government or operating in EU jurisdictions — will face growing pressure to produce compliant AI SBOMs. The guidance also omits a 'level of autonomy' element (flagged as deferred), signalling an open standards gap for agentic AI systems specifically.
Action needed
Adopt: begin gap analysis against the seven clusters; prioritise model lineage, dataset provenance, and security-properties elements as these are novel relative to conventional SBOMs. Vendor-risk teams should add AI SBOM attestation to procurement questionnaires immediately.