AI Bill of Materials (AI-BOM) Template

Document Type: Technical Standard / Template
Framework Alignment: SPDX 3.0, CycloneDX 1.6, OWASP AIBOM, NIST AI RMF 2.0, EU AI Act
Effective Date: 2026-01-15
Version: 1.0

---

1. Purpose and Scope

1.1 Purpose

This AI Bill of Materials (AI-BOM) template provides a comprehensive inventory framework for documenting all components, dependencies, and artifacts that comprise an AI system. The AI-BOM enables:

• Transparency: Complete visibility into AI system composition
• Traceability: Chain of custody from data to deployment
• Security: Vulnerability identification and supply chain risk management
• Compliance: EU AI Act, NIST AI RMF, and ISO/IEC 42001 alignment
• Auditability: Evidence for regulatory and internal audits

1.2 Scope

This template applies to:

• Internal AI/ML models (trained, fine-tuned, custom)
• Third-party AI components (APIs, SDKs, embedded models)
• Foundation models and GPAI (General Purpose AI)
• Training and evaluation datasets
• AI infrastructure and runtime environments

1.3 Regulatory Requirements

Regulation	AI-BOM Requirement	Reference
EU AI Act	Technical documentation for high-risk AI	Article 11
EU AI Act	Training data summary for GPAI	Article 53(1)(d)
NIST AI RMF 2.0	Model provenance and supply chain	MAP 1.5, MANAGE 2.3
ISO/IEC 42001	AI asset inventory	Clause 8.2, Annex A.5

---

2. AI-BOM Header Information

2.1 Document Metadata

Field	Value
AI-BOM ID	[Unique identifier, e.g., AIBOM-2026-001]
AI System Name	[System name]
Version	[AI-BOM version]
Created Date	[YYYY-MM-DD]
Last Updated	[YYYY-MM-DD]
Author	[Name/Team]
Owner	[System Owner]
Classification	[Public / Internal / Confidential / Restricted]
Risk Tier	[Low / Medium / High / Critical]

2.2 AI System Identification

Field	Value
System Registry ID	[Internal inventory ID]
System Type	[ ] Predictive/Classification [ ] Generative [ ] Recommender [ ] Computer Vision [ ] NLP [ ] Agentic
Deployment Status	[ ] Development [ ] Staging [ ] Production [ ] Deprecated
Primary Function	[Brief description of system purpose]
Business Domain	[Healthcare / Finance / HR / Operations / etc.]

---

3. Model Components

3.1 Primary Model(s)

Field	Model 1	Model 2	Model N
Model Name
Model ID/Version
Model Type
Architecture
Provider/Source	[ ] Internal [ ] Vendor [ ] Open-Source
Provider Name
License
Training Compute (FLOPS)
Parameter Count
Knowledge Cutoff
Download/Access Date
Hash/Checksum (SHA-256)
Verification Status	[ ] Verified [ ] Pending [ ] N/A

3.2 Model Provenance Chain

For each model, document the complete chain of custody:

┌─────────────────────────────────────────────────────────────┐
│ FOUNDATION MODEL                                            │
│ Name: [e.g., Llama-3-70B]                                  │
│ Provider: [e.g., Meta]                                     │
│ License: [e.g., Llama 3 Community License]                 │
│ Access Date: [YYYY-MM-DD]                                  │
│ Hash: [SHA-256]                                            │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│ FINE-TUNED MODEL                                            │
│ Name: [e.g., Company-Llama-Finance-v1]                     │
│ Fine-tuning Date: [YYYY-MM-DD]                             │
│ Fine-tuning Dataset: [Reference to Section 4]              │
│ Training Framework: [e.g., PyTorch 2.1, HuggingFace]       │
│ Hyperparameters: [Link to config file]                     │
│ Hash: [SHA-256]                                            │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│ DEPLOYED MODEL                                              │
│ Deployment ID: [Registry ID]                               │
│ Deployment Date: [YYYY-MM-DD]                              │
│ Environment: [Production/Staging]                          │
│ Endpoint: [API endpoint or service name]                   │
│ Version Tag: [e.g., v1.2.3]                                │
└─────────────────────────────────────────────────────────────┘

3.3 GPAI Model Classification (EU AI Act)

Applies to General Purpose AI models only:

Assessment Criteria	Value	Notes
Training Compute	[FLOPS value]	≥10²³ FLOPS = GPAI
Systemic Risk Threshold	[FLOPS value]	≥10²⁵ FLOPS = Systemic Risk
GPAI Classification	[ ] Not GPAI [ ] Standard GPAI [ ] Systemic Risk GPAI
AI Office Notification Required	[ ] Yes [ ] No	Required for systemic risk
Code of Practice Signatory	[ ] Yes [ ] No	Voluntary compliance path

---

4. Dataset Components

4.1 Training Datasets

Dataset ID	Name	Type	Source	License	Size	PII	Collection Date
DS-001		[ ] Proprietary [ ] Public [ ] Synthetic				[ ] Yes [ ] No
DS-002
DS-003

4.2 Dataset Provenance

For each training dataset, document:

Field	DS-001	DS-002
Original Source
Collection Method
Collection Date Range
Data Processing Applied
Anonymization Method
Quality Assessment
Bias Assessment
Right to Use Verification	[ ] Verified [ ] Pending
Copyright Compliance	[ ] Compliant [ ] Needs Review
robots.txt Respected	[ ] Yes [ ] No [ ] N/A

4.3 Training Data Summary (EU AI Act GPAI Requirement)

Required for GPAI models (Article 53(1)(d)):

Summary Element	Description
General Description	[High-level description of training corpus]
Data Categories	[Types of data included: text, images, code, etc.]
Data Sources	[Categories of sources: web crawl, licensed, proprietary]
Geographic Coverage	[Regions/languages represented]
Temporal Coverage	[Date range of source data]
Notable Exclusions	[What was explicitly excluded]
Synthetic Data Proportion	[% synthetic vs. natural data]
Copyright Approach	[How copyright compliance was achieved]

---

5. Software Dependencies

5.1 ML Frameworks and Libraries

Component	Version	License	Source	CVE Status	Last Checked
PyTorch
TensorFlow
HuggingFace Transformers
LangChain
scikit-learn
NumPy
Pandas
[Other]

5.2 Third-Party AI Services

Service	Provider	API Version	Contract ID	Data Processing	Retention Policy
				[ ] Processes [ ] Stores [ ] Trains

5.3 Infrastructure Components

Component	Provider	Version/SKU	Region	Security Certification
Compute
Storage
Model Serving
Vector DB
Monitoring

---

6. Runtime Configuration

6.1 Model Serving Configuration

Parameter	Value	Rationale
Max Tokens (Input)
Max Tokens (Output)
Temperature
Top-P
Frequency Penalty
Presence Penalty
System Prompt	[Link to version-controlled prompt]
Safety Filters	[ ] Enabled [ ] Disabled

6.2 Guardrails Configuration

Guardrail Type	Implementation	Threshold	Action
Input PII Detection
Input Injection Detection
Output Toxicity Filter
Output PII Redaction
Token Rate Limiting
Cost Rate Limiting

---

7. Agentic AI Components (If Applicable)

7.1 Agent Configuration

Field	Value
Agent Type	[ ] Single Agent [ ] Multi-Agent [ ] Orchestrated
Orchestrator	[If multi-agent, specify orchestrator]
Action Boundary	[Permitted action types]
Kill Switch	[ ] Implemented [ ] Tested [ ] Documented

7.2 Tool Access

Tool Name	Type	Access Level	Rate Limit	Audit Logging
	[ ] Read [ ] Write [ ] Execute			[ ] Yes [ ] No

7.3 Multi-Agent Communication

Agent ID	Role	Communication Protocol	Message Logging
			[ ] Yes [ ] No

---

8. Security and Vulnerability Assessment

8.1 Known Vulnerabilities

Component	CVE ID	Severity	Status	Mitigation
		[ ] Critical [ ] High [ ] Medium [ ] Low	[ ] Patched [ ] Mitigated [ ] Accepted

8.2 Security Assessments

Assessment Type	Date	Result	Report Link
Threat Model
Red Team Test
Penetration Test
Model Extraction Test
Data Leakage Test

8.3 Supply Chain Risk Assessment

Risk Category	Assessment	Mitigation
Model Poisoning	[ ] Low [ ] Medium [ ] High
Data Poisoning	[ ] Low [ ] Medium [ ] High
Dependency Vulnerability	[ ] Low [ ] Medium [ ] High
Third-Party Service Risk	[ ] Low [ ] Medium [ ] High
License Compliance Risk	[ ] Low [ ] Medium [ ] High

---

9. Compliance Attestations

9.1 Regulatory Compliance

Regulation	Applicable	Status	Evidence Location
EU AI Act (High-Risk)	[ ] Yes [ ] No	[ ] Compliant [ ] In Progress [ ] N/A
EU AI Act (GPAI)	[ ] Yes [ ] No	[ ] Compliant [ ] In Progress [ ] N/A
NIST AI RMF	[ ] Yes [ ] No	[ ] Aligned [ ] In Progress [ ] N/A
ISO/IEC 42001	[ ] Yes [ ] No	[ ] Certified [ ] In Progress [ ] N/A
FDA 21 CFR Part 11	[ ] Yes [ ] No	[ ] Validated [ ] In Progress [ ] N/A
HIPAA	[ ] Yes [ ] No	[ ] Compliant [ ] In Progress [ ] N/A
SOC 2	[ ] Yes [ ] No	[ ] Certified [ ] In Progress [ ] N/A

9.2 Internal Governance

Control	Status	Evidence
Inventory Registration	[ ] Complete
Risk Tiering	[ ] Complete
System Card	[ ] Complete
AIA (if High-Risk)	[ ] Complete [ ] N/A
Pre-Production Gate	[ ] Passed
Monitoring Active	[ ] Yes

---

10. Change History

Version	Date	Author	Changes
1.0			Initial AI-BOM

---

11. Approvals

Role	Name	Date	Signature
System Owner
Security Review
AI Governance

---

Appendix A: AI-BOM Generation Checklist

A.1 Minimum Required Fields (All AI Systems)

• AI-BOM ID and metadata
• Primary model identification
• Model provenance (origin, license)
• Training dataset sources
• Software dependencies with versions
• Security assessment status
• Governance approvals

A.2 Additional Fields (High-Risk AI)

• Complete model provenance chain
• Dataset provenance and bias assessment
• Full CVE assessment
• Red team results
• Compliance attestations

A.3 Additional Fields (GPAI)

• Training compute (FLOPS)
• Systemic risk classification
• Training data summary (public)
• Copyright compliance documentation
• AI Office notification (if applicable)

A.4 Additional Fields (Agentic AI)

• Agent configuration
• Tool access inventory
• Action boundaries
• Kill switch documentation
• Multi-agent communication map

---

Appendix B: SPDX 3.0 AI Profile Mapping

This AI-BOM aligns with SPDX 3.0 AI and Dataset profiles:

SPDX Field	AI-BOM Section
AI:modelType	Section 3.1
AI:trainingEnergy	Section 3.3 (compute)
AI:informationAboutTraining	Section 4.3
AI:limitation	Section 3.2 (System Card)
Dataset:datasetType	Section 4.1
Dataset:dataCollectionProcess	Section 4.2
Dataset:intendedUse	Section 4.2

---

Document Classification: [Internal / Confidential]
Next Review Date: 2027-01-15

---

CODITECT AI Risk Management Framework

Document ID: AI-RMF-13 | Version: 2.0.0 | Status: Active

---

AZ1.AI Inc. | CODITECT Platform

Framework Alignment: NIST AI RMF 2.0 | EU AI Act | ISO/IEC 42001

---

This document is part of the CODITECT AI Risk Management Framework. For questions or updates, contact the AI Governance Office.

Repository: coditect-ai-risk-management-framework Last Updated: 2026-01-15 Owner: AZ1.AI Inc. | Lead: Hal Casteel