SBOM and MBOM Architecture Guidelines

Document Reference: AUC-TECH-BOM-001
Version: 2.2.0
Effective Date: October 24, 2026
Last Updated: October 24, 2026
Classification: Technical Standard / Engineering Reference
Standards Alignment: Executive Order 14028 | NTIA Minimum Elements | CycloneDX v1.6 | SPDX 3.0 | NIST AI RMF

Executive Summary (ISO 24495-1 Plain-Language Overview)

[!IMPORTANT] Legal Precedence Notice: This plain-language summary is provided solely for reader convenience and technical audit navigation pursuant to ISO 24495-1:2023. It reflects engineering supply chain standards. In the event of any conflict, governing contractual security commitments and operative policy articles strictly prevail.

• Dual-BOM Architecture: Auctoric establishes end-to-end supply chain transparency across traditional code (SBOM) and generative machine learning models (MBOM).
• Open Industry Standards: Standardizes on CycloneDX v1.6 JSON and SPDX 3.0 JSON, satisfying NTIA Minimum Elements under Executive Order 14028.
• AI Model Transparency (MBOM): Tracks model architecture, parameter scale, SHA-256 tensor hashes (Safetensors only; Pickle banned), dataset provenance, and benchmark evaluations.
• Cryptographic Attestations: Builds are hermetically produced, signed with Sigstore Cosign, and embedded with in-toto build provenance meeting SLSA Level 3 criteria.
• Verification & Delivery: Customers and auditors can verify container digests using public Cosign keys or download inventory files via Trust Center API.

High-Intent Supply Chain FAQ (AEO Direct Answers)

What formats does Auctoric use for Software and Model Bills of Materials?

Under Section 3 and Section 4, Auctoric generates SBOMs and MBOMs in CycloneDX v1.6 JSON (primary) and SPDX 3.0 JSON (secondary), adhering to NTIA Minimum Elements under U.S. Executive Order 14028.

How does an MBOM differ from a standard software SBOM in Auctoric’s architecture?

Under Section 4, while an SBOM catalogs application dependencies, an MBOM captures AI-specific supply chain metadata: model architecture, parameter count, SHA-256 tensor hashes (Safetensors only), dataset lineage, quantization formats, and safety benchmark scores.

How are Auctoric software builds and model artifacts cryptographically verified?

Under Section 5, all build artifacts are cryptographically signed during CI using Sigstore Cosign with keyless OIDC signatures and in-toto provenance attestations, verifiable by customers against Auctoric’s public key.

1. Overview and Rationale

In modern cloud-native and AI-driven platforms, supply chain transparency is a core security prerequisite. Auctoric, LLC (“Auctoric”) enforces supply chain visibility across both classical software dependencies and artificial intelligence models through:

1. Software Bill of Materials (SBOM): Machine-readable inventory of software components, third-party libraries, container layers, and licenses.
2. Model Bill of Materials (MBOM): Structured inventory of machine learning models, weights, training dataset provenance, tokenizers, fine-tuning checkpoints, and evaluation benchmarks.

These guidelines define the generation, signing, validation, and publication standards for Auctoric deliverables.

2. Supply Chain Architecture & Lifecycle

flowchart LR
    subgraph BuildPipeline["Auctoric Automated Build Pipeline"]
        SourceCode["Code Repo & ML Pipeline"] --> Build["Hermetic Container Build"]
        Build --> Syft["SBOM Generator (CycloneDX)"]
        Build --> MBOMLog["MBOM Extractor (Model Metadata)"]
        Syft --> Cosign["Sigstore Cosign Signing"]
        MBOMLog --> Cosign
        Cosign --> Attest["In-Toto Cryptographic Attestation"]
    end

    subgraph Registries["Secure Artifact Registries"]
        Attest --> OCI["OCI-Compliant Authenticated Container Registry"]
        Attest --> ModelReg["Encrypted Object Storage / Model Registry"]
    end

    subgraph Consumers["Downstream Verification"]
        OCI --> Verify["Customer / Auditor Verification via Cosign"]
    end

3. Software Bill of Materials (SBOM) Specification

3.1 Standard & Schema Format

• Mandatory Format: CycloneDX v1.6 JSON (Primary) or SPDX 3.0 JSON (Secondary upon federal requirement).
• Compliance Alignment: Fulfills the National Telecommunications and Information Administration (NTIA) Minimum Elements for an SBOM under Executive Order 14028.

3.2 Required Metadata Fields

Every generated SBOM must include:

• Author / Supplier: Auctoric, LLC
• Component Name & Version: Specific semantic release (e.g., auctoric-agent-runtime:v2.4.1)
• Package URL (PURL): Canonical identifier (e.g., pkg:golang/github.com/gin-gonic/gin@v1.9.1)
• Cryptographic Hashes: SHA-256 for all binaries and dependency files
• Dependency Relationships: Direct vs transitive dependency tree
• License Information: SPDX license expression (e.g., Apache-2.0, MIT)

4. Model Bill of Materials (MBOM) Specification

Artificial intelligence models introduce unique supply chain risks, such as backdoors, data poisoning, and licensing contagion. Auctoric extends CycloneDX machine learning schema standards to produce verifiable MBOMs.

4.1 Required MBOM Metadata Fields

4.2 Example CycloneDX v1.6 MBOM JSON Snippet

{
  "$schema": "http://cyclonedx.org/schema/bom-1.6.schema.json",
  "bomFormat": "CycloneDX",
  "specVersion": "1.6",
  "serialNumber": "urn:uuid:3e671687-395b-41f5-a30f-a58921a69b79",
  "version": 1,
  "metadata": {
    "timestamp": "2026-10-22T08:00:00Z",
    "component": {
      "type": "machine-learning-model",
      "name": "auctoric-governance-evaluator-70b",
      "version": "v1.4.0",
      "hashes": [
        {
          "alg": "SHA-256",
          "content": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
        }
      ],
      "modelCard": {
        "modelParameters": {
          "approach": { "type": "supervised" },
          "task": "classification-and-reasoning",
          "architectureFamily": "Transformer-Decoder",
          "parameters": 70000000000
        },
        "quantitativeAnalysis": {
          "performanceMetrics": [
            { "type": "toxicity-benchmark", "value": "0.001" },
            { "type": "prompt-injection-resistance", "value": "99.4%" }
          ]
        }
      }
    }
  }
}

5. Build Pipeline Automation and Verification

1. Automated Generation: Generated during CI release builds using syft and Auctoric’s mbom-extractor.
2. Cryptographic Signing: Signed using Sigstore Cosign with automated keyless OIDC signing tied to Auctoric’s corporate GitHub organization.
3. In-Toto Attestations: Build provenance attestations (SLSA Level 3 criteria) stored directly in the OCI registry alongside container artifacts.
4. Customer Verification: Enterprise customers can verify release integrity using the public Cosign verification key:

   cosign verify --key https://auctoric.ai/security/cosign.pub registry.auctoric.ai/platform/engine:v2.4

6. Access and Customer Distribution

Enterprise customers and auditors may access current SBOM and MBOM inventories:

• Via API: GET /api/v1/compliance/bom?artifact_id={ID}
• Via Portal: Downloadable through the Auctoric Trust Portal (https://trust.auctoric.ai/bom)
• Security Inquiries: security@auctoric.ai