AI System Inventory

How to use this document

• What it is for. This is the single register of all of the organization's AI: tools used by staff, systems it integrates or sells, agents that act in its systems and models it hosts, trains or fine-tunes. For each entry it documents purpose, owners, provenance, hosting, data, risk level, version, status and controls. Clients and auditors often ask for it at the start of a review.
• Single register. Other documents do not keep their own inventories: they point to this one. The technical detail of each third-party or open-weight model is documented in the "Third-Party and Open-Weight Model Provenance Sheet", and the technical life cycle of models and datasets under the "AI Model and Dataset Inventory Policy"; both are linked to the entry in this inventory.
• Who should own it. The AI lead, supported by information security. Each entry also has a system owner who maintains it. Enter the function in [RESPONSIBLE FUNCTION] and the person in [OWNER].
• Decisions to take first. Decide where the inventory is kept (controlled spreadsheet, asset management or governance tool), which deployment profiles the organization has ("Module selection" section), who approves additions and removals in each approval lane, and how it connects with procurement and IT.
• Fields to complete. Complete the summary view (Section 6) for every entry; the core record (Sections 5.1 and 5.2) for every medium- or high-risk entry; and the module blocks (5.3 to 5.5) according to the entry type.
• Country variants. Keep the European Union, Spain, Peru, Ecuador or United States variants that correspond to where you operate, and add the fields they indicate to the record. Delete the rest.
• Review. Ask Legal or your counsel to confirm which fields and retention periods the laws that apply to you provide for; the [COMPLETE] fields in the variants mark points worth confirming.
• Approval and use. Have the AI committee (or management, in a small organization) approve the inventory structure, record the approval in "Version history" and tell all areas that no tool, system, agent or model moves to pilot or production without being registered.

Module selection

Profile or level	Sections to keep
All	Sections 1 to 4, 5.1, 5.2, 6 to 11
P1: uses AI inside software it buys	Core; in 5.2, developer, country, hosting mode and account terms are enough
P2: integrates models through APIs or RAG	Core, with all fields in 5.2
P3: hosts models on its own or rented infrastructure	Core and 5.3
P4: trains or fine-tunes models	Core and 5.4
P5: uses agents that act in systems	Core and 5.5
Basic level	Summary view for every entry; core record only for medium- or high-risk entries
Standard and Reinforced levels	Summary view and full record with the applicable modules

Profiles are cumulative: an organization can be P1, P2 and P5 at the same time. Delete the modules that do not apply.

AI System Inventory

Field	Value
Organization	[LEGAL ENTITY NAME]
Owning function	[RESPONSIBLE FUNCTION]
Inventory owner	[OWNER]
Approved by	[APPROVED BY]
Effective date	[DATE]
Review frequency of the full inventory	[REVIEW FREQUENCY]
Inventory location	[COMPLETE]
Contact	[AI GOVERNANCE EMAIL]

1. Purpose

1.1 This inventory identifies all AI of [ORGANIZATION NAME] and documents, for each entry, its purpose, owners, provenance, hosting, functioning, data, impacts, risk level, version, status and controls.

1.2 It is used to: know and control the use of AI; detect unauthorized and duplicate uses; know where each model comes from and where it runs; prioritize assessments, oversight and audits; respond to clients, auditors and authorities; and connect each entry to its assessments, contracts, incidents and decisions.

2. What is registered

2.1 All AI used with the organization's information or in its processes is registered. Each entry has one of these types:

Entry type	What it includes	Examples
Tool	AI application or feature used by staff, including AI features enabled in software already under contract	Corporate chat assistant, office copilot, meeting transcription
System	Solution that combines one or more models with data, rules and interfaces for a specific purpose, in-house or third-party, including those the organization sells	Customer service chatbot, scoring engine, RAG search
Agent	System that plans and executes actions in other systems with some degree of autonomy	Agent that sends emails, handles tickets, buys or runs code
Model	AI model the organization consumes through an API, hosts, trains or fine-tunes, when managed as an asset of its own	Language model via API, self-hosted open-weight model, fine-tuned model

2.2 Pilots and tests that use real data or processes are also registered.

2.3 When several components, models or agents together produce an output, one system is registered and its components are listed; each model or agent that has its own owner, version or permissions also has its own entry, linked to the system.

2.4 Section 7 records rejected use cases and unauthorized uses detected, with their resolution.