Originally published at https://charz.ai/blog/ai-system-inventory-vs-ai-risk-register by Char-Z AI.
What You Have Versus What Could Go Wrong
An AI system inventory is the record of the AI systems you operate: what each one does, who owns it, what model and vendors it depends on, what data it touches, where it is deployed, and who can use it. It answers a factual question — what do we have?
An AI risk register is the record of the risks those systems carry: what could go wrong, how likely and severe it is, what you are doing about it, and who is accountable. It answers a judgement question — what could go wrong, and how much do we care?
The two are related but they are not the same document, and merging them is one of the most common causes of both going stale. An inventory changes when systems, ownership, models, or data change. A register changes when risk judgements change — and it also changes when the inventory changes, which is why the link between them matters.
Side by Side
- Question answered — inventory: what do we operate? Register: what could go wrong, and what are we doing about it?
- Unit of record — inventory: one AI system. Register: one risk.
- Nature — inventory: factual. Register: judgemental.
- Typical fields — inventory: name, purpose, owner, model, vendors, data categories, deployment, users, status. Register: risk description, category, likelihood, impact, score, mitigation, owner, status, due date, last reviewed.
- Change trigger — inventory: system, model, vendor, data, or ownership change. Register: new assessment, incident, control change, or inventory change.
- Primary audience — inventory: procurement, security, architecture, governance. Register: leadership, auditors, risk owners, governance.
- Goes stale because — inventory: nobody assigned to maintain it. Register: no review cadence.
- Cardinality — inventory: one row per system. Register: many rows per system.
What Belongs in Each
An inventory entry needs enough context to answer the questions a reviewer will ask without a follow-up meeting. At minimum: what the system is called, what it does, what it is for, who owns it, which models it depends on, which vendors are involved, what categories of data it processes, where it is deployed, who can access it, and whether it is in production or retired.
A register entry needs enough structure to compare and prioritise risks across systems: a stable identifier, the system it attaches to, a description of what could go wrong, a category, likelihood and impact on a consistent scale, a combined score, the treatment, the accountable owner, a status, a due date, and the date it was last reviewed.
Notice the asymmetry. The inventory is descriptive; the register is evaluative. The link between them is the system identifier, and it is the single field that makes both records useful together — it is how you answer "what risks attach to this system?" and "which systems carry no assessed risk at all?"
How They Feed Each Other
- Inventory to register — a new system, vendor, or data category seeds new risk entries.
- Register to inventory — a risk about the same system across several entries reveals an inventory gap.
- Both to everything else — vendors, data flows, policies, and assessments all key off the system record.
- Both from operations — incidents generate risk entries and often update the inventory.
The feedback loop is what keeps the pair honest. A register with risks against a system that no longer exists is a signal the inventory is not maintained. An inventory with systems that have no risk entries is a signal the register is not being seeded.
Keeping Both Current
Own them separately. One named owner for the inventory, another for the register cadence, even if the same person holds both roles in a small team. Ambiguous ownership is how both records decay.
Review on different clocks. The inventory is checked whenever something changes and refreshed on a regular pass. The register is reviewed on a fixed cadence — monthly for open treatments, annually for a full re-score — plus on change.
Make the link explicit. A register entry that does not name its system cannot be queried, prioritised, or retired. Enforce the system reference at entry.
Record the review, not just the conclusion. The value of a register is the loop around it: what was reviewed, when, and what changed. Without that record, the register is a snapshot, not a process (NIST, 2023).
Use one tool, not two files. This is where spreadsheets stop working well. Two linked registers maintained by hand diverge the moment one is updated without the other. A structured store with the system reference enforced keeps them consistent by construction.
Frequently Asked Questions
Can one spreadsheet hold both the inventory and the risk register?
It can for a handful of systems, if the system reference is enforced consistently. It breaks down when the same system needs several risk entries and when history matters — the point at which a structured record earns its keep.
Which should I build first?
The inventory. Risks attach to systems, so the systems must exist as records first. An assessment performed before the inventory exists will spend its early hours producing one.
Should the inventory include retired systems?
Yes, with a status field. Retired systems still hold data, still appear in historical decisions, and still matter for retention and deletion. Deleting the row loses that context.
How many risk entries should a system have?
As many as it genuinely carries. Forcing one entry per system hides compound risks; a single generic entry per system is a sign the assessment was done for form rather than for judgement.
Sources
NIST. (2023). *Artificial Intelligence Risk Management Framework (AI RMF 1.0)*. National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.100-1
European Commission. (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council. *Official Journal of the European Union*. https://eur-lex.europa.eu/eli/reg/2024/1689
Top comments (0)