Unapproved AI use
Objective: protect client trust. Driver: staff pasting client material into public AI tools. Likelihood high, impact high. Safeguard: an approved-tools list and a clear paste rule. Status: open, owner COO, quarterly.
The AI risk register
What goes in it, with a finished example.
The structure to copy, four worked entries, and what a completed AI risk picture looks like. Built for your business register, not another silo.
TL;DR
- AI risks belong in your main risk register, held against your business objectives, not in a standalone AI register.
- The full structure is on this page to copy: the standard register fields plus three AI-specific additions.
- Four worked entries show what good looks like.
- Instead of a blank template, see what finished looks like: a completed AI risk picture for an anonymised business.
- Free, no form.
What this guide covers:
What is an AI risk register?
People build two different things under this name. One is a standalone register of AI-system risks: the tools you have adopted, their data, their accuracy. Most templates you will find are that. The other treats AI as entries in your main business risk register, held against your objectives, and that is what this page is for, because AI is not a new risk on your register; it is the largest single driver of the risks already there.
A standalone AI register answers what could go wrong with our AI. Your business register answers the question your board actually asks: what could stop us achieving what we set out to do this year, and how much of that is AI driving?
The template
The structure: what every entry carries
The standard register fields first: a risk ID, the risk described plainly, likelihood, impact, a rating, an owner, safeguards, status, and a review date. Any register you already keep has most of these.
Then the three AI-specific additions that make it an AI risk register:
- The objective it threatens. Every entry hangs off something the business is trying to achieve. This is the field that turns a list into a decision.
- The AI driver. What AI actually changed: speed, cost, or who can plausibly do it. If nothing moved, it does not need this field filled.
- Both directions. A flag for whether the same shift creates an opening as well as a threat. Most do.
Copy this structure into whatever you keep your risks in. It is deliberately compatible with a conventional register: extend, do not rebuild.
Example entries: what good looks like
Invoice fraud, rewritten
Objective: protect margin. Driver: voice cloning defeats phone verification of bank-detail changes. Was low, now high. Safeguard: dual-channel verification, no voice-only approval. Owner CFO, quarterly.
AI answers reaching customers
Objective: grow retention. Driver: confident wrong answers from AI-assisted support. Likelihood medium, impact high. Safeguard: human review on customer-facing AI output. Owner Head of CX, quarterly.
Competitor response speed
Objective: win new customers. Driver: smaller rivals answering at a speed that used to need a bigger team. Both directions: the same capability is open to us. Owner CEO, quarterly.
Instead of a blank grid
A template shows you the boxes. This shows you the finished shape.
Templates get downloaded to answer one question: what should this look like when it is done well? A blank grid is the least useful answer. So rather than another empty template, the structure above is the finished shape: the entries, the scores, the safeguards, and the report a leadership team would see.
Drova’s AI Disruption Index builds exactly that from your objectives, in about ten minutes, with your data rather than an example.
One register, not two
Why AI risks belong in your main register
A standalone AI register feels tidy and usually goes the same way: it is owned by nobody in particular, reviewed after everything else, and read by no one making decisions. The risks in it are real; the document is a silo.
Held in your main register against your objectives, the same entries compete for attention with everything else that threatens the plan, which is exactly the comparison a leadership team needs. AI is not a category of risk to file separately. It is a force moving the risks you already rank.
How Drova helps
If you want it built for you
The structure above is yours to use anywhere. If you would rather start from filled-in than blank: Drova's AI Disruption Index builds the register entries from your objectives, scores how hard AI is driving each risk, drafts a safeguard for every one it raises, and hands you the report to take into your next leadership conversation. About ten minutes, free, and what comes back is your picture, not your industry's.
FAQs
AI risk register FAQs
Where is the downloadable template?
The structure is on this page to copy: the standard register fields plus the three AI-specific additions. We deliberately put the finished shape on the page rather than hand over a blank grid, because a blank grid is the part you can already draw yourself. If you would rather start from filled-in, the Index builds the entries from your objectives.
Do I need a separate AI risk register?
We would argue no. A standalone AI register tends to become a silo nobody owns. AI risks belong in your main register, held against your objectives, competing for attention with everything else that threatens the plan.
What fields does an AI risk register need?
The standard set: risk ID, description, likelihood, impact, rating, owner, safeguards, status, review date. Plus three AI-specific fields: the objective the risk threatens, the AI driver behind it, and whether the same shift also creates an opening.
How is the register different from an AI risk assessment?
The assessment is the exercise: working out what AI has changed and what the scores should be. The register is the standing record those results live in. Run the assessment, record it in the register, review on a schedule.
Who should own it?
The register as a whole belongs to whoever owns risk in the business. Each entry belongs to whoever owns the objective it threatens, which is usually not the same person.
Should it map to NIST or ISO 42001?
It can later, if a customer or regulator asks. The structure on this page does not require any framework to be useful, and starting is more valuable than mapping.
Your register entries, built from your objectives, with a safeguard drafted for each. Free.
