How to Build an AI Risk Register: What to Record and Who Owns It
An AI risk register is the working record of what could go wrong with each AI system, how serious it is, who owns it and what is being done. NIST's AI Risk Management Framework does not use the term, but it asks organisations to document exactly the things a register holds. Here are the fields, the owners, the review rhythm and the common failures.
Key Takeaways
- Start from the AI system inventory. A register records risks against systems, so it can only be as complete as the list of systems it is built on.
- NIST does not prescribe a 'risk register', but it asks you to document what one holds: organisational risk tolerances (MAP 1.5), the likelihood and magnitude of each identified impact (MAP 5.1), prioritised treatment and responses (MANAGE 1.2 and 1.3), and negative residual risks (MANAGE 1.4).
- Every entry needs a named owner, and the register itself needs an owner, because NIST asks for roles and responsibilities for mapping, measuring and managing AI risks to be documented and clear (GOVERN 2.1).
- Record what you cannot measure, and what you decided to accept. NIST asks for risks that will not or cannot be measured to be documented (MEASURE 1.1), and lists accepting a risk as one of four response options alongside mitigating, transferring and avoiding.
- Review on a schedule and on triggers. NIST asks for ongoing monitoring and periodic review with a defined frequency (GOVERN 1.5) and for tracking unanticipated and emergent risks over time (MEASURE 3.1).
When a board asks "what are the risks of the AI we use?", the answer should not be a conversation. It should be a document: for each system, what could go wrong, how serious it would be, who owns it, and what has been decided. That document is an AI risk register.
This piece sets out what to put in one, who should own each part and how to keep it alive. It is advisory. It describes the artefact and the decisions around it, and anchors them in the NIST AI Risk Management Framework. It should be read alongside our piece on the AI system inventory, which comes first.
A Note on the Term
NIST's AI Risk Management Framework (AI RMF 1.0, January 2023) does not use the phrase "risk register". We checked the framework text and the Generative AI Profile (NIST AI 600-1, July 2024). What the framework does is ask organisations to document a series of things about risk. A register is a practical way to hold them in one place. We use the term as shorthand for that documentation, not as a NIST requirement.
The NIST Playbook page states that AI RMF 1.0 is being updated, so check the current text when you design your own.
Start From the Inventory
A register records risks against AI systems. If a system is not on your inventory, it has no entry, and the register will look complete while missing the systems you know least about. Build or confirm the inventory first, then give each system a register section. Where several systems share a model or a vendor, record the shared risk once and link it.
What NIST Asks You to Document
The AI RMF defines risk as the composite measure of an event's probability of occurring and the magnitude or degree of the consequences. Several subcategories ask for specific things to be documented. Together they describe the contents of a register:
- MAP 1.5 — Organisational risk tolerances are determined and documented. Register field it implies: A statement of tolerance the ratings are judged against
- MAP 5.1 — Likelihood and magnitude of each identified impact, beneficial and harmful, are identified and documented. Register field it implies: Likelihood and impact, per risk
- MEASURE 1.1 — Risks that will not or cannot be measured are properly documented. Register field it implies: A "not measured" flag with the reason
- MANAGE 1.2 — Treatment of documented AI risks is prioritised based on impact, likelihood and available resources. Register field it implies: A priority and a treatment decision
- MANAGE 1.3 — Responses to high-priority risks are developed, planned and documented; options include mitigating, transferring, avoiding or accepting. Register field it implies: A chosen response and an action plan
- MANAGE 1.4 — Negative residual risks to downstream acquirers and end users are documented. Register field it implies: A residual risk statement after treatment
- MEASURE 3.1 — Approaches are in place to regularly identify and track existing, unanticipated and emergent AI risks. Register field it implies: Review dates and triggers
- GOVERN 2.1 — Roles, responsibilities and lines of communication for mapping, measuring and managing AI risks are documented and clear. Register field it implies: Named owners
NIST describes risk tolerance as the organisation's readiness to bear risk to achieve its objectives, notes that the framework does not prescribe a tolerance, and says tolerances are contextual and likely to change over time. That is why the first field matters: a rating means little until you have said what level you are prepared to accept. NIST also describes residual risk as the risk remaining after risk treatment (a definition it takes from ISO Guide 73).
The Fields We Suggest
The list below is our suggested layout, built from the NIST items above. It is a recommendation, not a NIST template. Keep it small enough that people will fill it in.
- Risk ID — A unique reference
- System — Link to the inventory record
- Risk description — A specific statement: what happens, to whom, in what circumstances. "Bias" is not a risk; "the screening tool ranks one group of applicants lower for reasons unrelated to the job" is
- Source — Where the risk came from: design, data, vendor, use, misuse, change over time
- Who is affected — Customers, staff, third parties, the organisation
- Likelihood and impact — Each on a scale you have defined in words, with the reasoning
- Measurable? — Whether you can measure it, and how. If you cannot, say so and why
- Existing controls — What is already in place
- Response — Mitigate, transfer, avoid or accept, and the plan
- Residual risk — What remains after the response, and who has accepted it
- Risk owner — One named person accountable for this entry
- Review date and triggers — When it is next reviewed and what would prompt an earlier review
- Status — Open, being treated, accepted, closed
An illustrative entry
The example below is invented to show the level of detail. The system and the figures do not describe any real organisation.
System: an invented "customer-reply drafting assistant" used by a support team. Risk: the assistant drafts a confident but incorrect statement about a refund entitlement, and an agent sends it without checking. Affected: customers, and the organisation (complaints, possible refunds owed). Likelihood / impact: medium / medium, reasoning recorded. Measurable? Partly: a monthly sample of sent replies can be checked against policy; the rate at which a reviewer would miss an error cannot be measured directly, and that is recorded. Response: mitigate (second-person review for any reply mentioning money). Residual risk: some incorrect replies will still be sent; accepted by the head of support. Owner: head of support. Review: quarterly, and on any change to the underlying model.
Notice what makes the entry useful: it is specific, it names a person, it records what cannot be measured, and it states who accepted what remains.
Who Owns What
Three kinds of owner are involved. Write each one down.
- The register owner, usually a governance, risk or compliance lead, who keeps the structure, chases reviews and reports on it. NIST's GOVERN 2.1 asks for roles and responsibilities to be documented and clear.
- Each risk owner, the person accountable for the entry, normally the business owner of the system and not the security or governance team. Governance can challenge a rating, but the person who runs the system is best placed to know it.
- The risk acceptor. Accepting a risk is a decision about the organisation's tolerance. Decide in advance which level of risk needs which level of seniority to accept, and record who accepted each one.
Rating Risk Without False Precision
NIST's definition of risk combines likelihood and magnitude, and says nothing about how to score them. A short scale defined in words (for example low, medium and high, each with a one-line description) is usually enough, and is easier to apply consistently than a numeric formula. Whatever you choose, write the definitions down, apply them the same way to every entry, and resist turning judgement into precise-looking numbers. NIST's own guidance on measurement is to start with the most significant risks and to document those that cannot be measured.
Keeping the Register Alive
NIST's GOVERN 1.5 asks for ongoing monitoring and periodic review of the risk management process and its outcomes, with roles defined and the frequency of review determined. MEASURE 3.1 asks for approaches to track existing, unanticipated and emergent risks. In practice, that means two review rhythms:
- Scheduled. A set cycle for the whole register, and a more frequent one for high-priority entries.
- Triggered. An earlier review when something changes: a new use of the system, a change to the underlying model or vendor, an incident, new findings from red teaming, or a change in the rules that apply to you.
MANAGE 4.3 adds that incidents and errors are communicated to relevant parties and that processes for tracking, responding to and recovering from them are followed and documented. Your incident response process should feed the register, so that each incident updates a risk rather than disappearing into a ticket.
Where Registers Fail
- Generic entries. "Data privacy risk" cannot be owned, rated or treated. Write the specific scenario.
- No tolerance statement. Without one, "accept" and "mitigate" are arbitrary decisions.
- No owner, or the wrong owner. An entry owned by "the AI team" is owned by nobody.
- Unmeasured risks left out. The hardest risks to measure are often the ones that matter; NIST asks for them to be documented.
- No residual risk. Treatment is not the end. State what remains and who accepted it.
- A register built once. It goes stale as systems and uses change. Give it a review rhythm and triggers.
- Disconnected from the rest. A register that does not link to the inventory, the system documentation and the incident process is a spreadsheet, not a control. If you follow the NIST AI RMF, our plain-English guide to the framework shows where the register fits.
What Leaders Should Be Asking
- Is there one register that covers every AI system on our inventory?
- Have we written down our risk tolerance, and who approved it?
- Does each entry have one named owner, and has it been reviewed within its review period?
- Which risks have we decided to accept, and who accepted them?
- Which risks can we not measure, and have we said so?
- What would make us review an entry early, and has that happened this year?
The Underlying Point
A risk register does not reduce risk by existing. It makes risk visible, owned and decided. The value is in the discipline of writing specific risks, naming the people accountable and revisiting them as systems change, which is the same documented, continuing management that NIST's framework asks for.
Imagine Works advises enterprise leaders on AI governance and risk design, including how to scope an AI risk register, set a risk tolerance statement and decide who owns and accepts each risk. We advise; we do not build or implement software. Get in touch to discuss your AI governance programme.
Sources (as of 16 November 2026)
All opened and read on 16 November 2026.
- 1NIST AI 100-1, Artificial Intelligence Risk Management Framework (AI RMF 1.0), January 2023 (risk definition, section 1.2.2 risk tolerance, residual risk, MAP 1.5, MAP 5.1, MEASURE 1.1, MEASURE 3.1, MANAGE 1.2 to 1.4, MANAGE 4.3, GOVERN 1.5, GOVERN 2.1)
- 2NIST AI 600-1, Generative Artificial Intelligence Profile, July 2024 (searched for "risk register": no occurrence)
- 3NIST AI RMF Playbook (GOVERN) (notes that AI RMF 1.0 is being updated and the Playbook will follow)
Related Service
AI Governance & Risk Design
Designing the governance framework and risk architecture that keeps your AI systems compliant, auditable, and board-ready — before regulation forces the issue.
Explore this serviceMore Insights
More on AI Governance
The NIST AI Risk Management Framework: What Enterprise Leaders Need to Understand
The NIST AI Risk Management Framework (AI RMF 1.0) has quietly become the most widely adopted enterprise AI governance framework in the United States — 57–67% of CISOs use it according to the 2026 Hitch Partners Global CISO Leadership Report. Voluntary, but load-bearing. Here is what the framework contains, how the 2024 Generative AI Profile extends it, and why it endured a change of US administration when the executive order that popularised it did not.
Generative AI Acceptable-Use Policy: What to Include
A generative AI acceptable-use policy is the control that answers shadow AI: it tells staff which tools they may use, what data may go in, how outputs must be checked and what is off limits. Here is a section-by-section checklist of what to include, anchored in NIST's generative AI profile and OWASP's guidance on sensitive information disclosure.
When Agents Go Rogue: Containment Controls Boards Should Ask About
An AI agent does not just answer — it acts. So its failure mode is a wrong action in the real world, at machine speed. Boards do not need to understand the model to govern this; they need to ask a small number of containment questions and recognise a good answer from a reassuring one. Here are the six, and how they map to the EU AI Act, ISO/IEC 42001, and the NIST AI Risk Management Framework.