What should a small business AI risk register record?

By Hugh Donatello, LAN Service Group · Published October 20, 2026 · Updated October 20, 2026

✓ Checked against Microsoft's Data, Privacy, and Security for Microsoft 365 Copilot page on October 19, 2026 (the Copilot example row). No other vendor terms or prices are stated on this page.

Short answer

One row per AI use, not per vendor. Each row records the use case, a named owner, the data the tool touches, what could go wrong, a likelihood and impact rating, the control you chose, any risk you accept and who accepted it, and the next review date. Those columns line up with outcomes in NIST's AI Risk Management Framework, across Govern, Map, Measure and Manage. CIO Support's free Excel register has the columns, six worked example rows and the NIST mapping.

Free AI risk register template

One row per AI use: owner, data touched, risk, likelihood and impact with the rating worked out for you, control, residual risk and review dates, six worked example rows and a NIST AI RMF mapping tab. No form or sign-up. General guidance, not legal advice; not published or endorsed by NIST.

Download Excel (.xlsx)

What an AI risk register is, and what it is not

An AI risk register is a working list of the ways your company uses AI, what could go wrong with each use, and what you are doing about it. It sits between two documents a small business may already have. An approved-tools list says which AI tools staff may use and with what data. An AI acceptable-use policy says how they must use them. The register is where you write down, use by use, the risks those rules are meant to control and whether the controls are in place.

The register is organized by use, not by tool, because risk depends on the use. NIST AI 100-1, the AI Risk Management Framework, defines risk as a composite measure of an event's probability of occurring and the magnitude or degree of its consequences. The same chatbot can be low risk when it tidies an internal memo and high risk when it writes a quote a customer will rely on, so each of those gets its own row.

NIST does not publish an AI risk register template as part of the AI RMF, and the framework does not require one. NIST's closest format is the general cybersecurity risk register template published with NIST IR 8286 Rev. 1, Integrating Cybersecurity and Enterprise Risk Management, which is not AI-specific. The AI RMF's outcomes do ask for the things a register records. GOVERN 1.6 asks for mechanisms to inventory AI systems. MAP 5.1 asks that the likelihood and magnitude of each identified impact are identified and documented. MANAGE 1.2 asks that treatment of documented risks is prioritized based on impact, likelihood and available resources or methods, and MANAGE 1.4 asks that negative residual risks are documented. For a company of ten to fifty people, a spreadsheet is usually the simplest way to keep all of that in one place.

The columns, and the NIST outcome each one supports

Each column below answers one question an owner, auditor, customer or insurer is likely to ask about an AI use. The right-hand column names the AI RMF subcategory the column helps you show, so the register can be traced back to NIST AI 100-1.

AI risk register columns for a small business, mapped to NIST AI RMF outcomes
Column	What to write	NIST AI RMF outcome it supports	Source
Use case	The tool and what it is used for, in one line, such as 'ChatGPT to draft customer emails'. One row per use.	GOVERN 1.6 inventory of AI systems; MAP 1.1 intended purposes and context of use are understood and documented	[1]
Owner	One named person who answers for this use, approves changes to it and keeps the row current.	GOVERN 2.1 roles and responsibilities are documented and clear; GOVERN 2.3 executive leadership takes responsibility for AI risk decisions	[1]
Data touched	The most sensitive data that goes into the tool or that it can reach, using your own data levels, such as Public, Internal, Confidential and Restricted. Treat protected health information (PHI) and GxP records as Restricted.	MAP 1.1 context of use; MEASURE 2.10 privacy risk is examined and documented	[1]
Who is affected	The people a wrong or leaked output could hurt: customers, applicants, employees, patients or the company itself.	MAP 1.1 potential positive and negative impacts to individuals, communities and organizations	[1]
Risk	What could go wrong, in plain words, and the generative AI risk it falls under, such as confabulation or data privacy.	MAP 5.1 each identified impact is documented; NIST AI 600-1 lists 12 generative AI risks	[1] [4]
Likelihood and impact	A rating for each on a scale you apply the same way to every row, and the score they make together.	MAP 5.1 likelihood and magnitude of each identified impact; MANAGE 1.2 treatment is prioritized based on impact and likelihood	[1] [2]
Control and response	What you do about it, and which response it is; NIST AI 100-1 names mitigating, transferring, avoiding or accepting as options.	MANAGE 1.3 responses to high-priority risks are developed, planned and documented (CIO Support suggests recording a response on every row)	[1] [2]
Test or check done	What you tried before relying on the use or its control, the date, and anything you could not measure.	MEASURE 1.1 start with the most significant risks and document what cannot be measured; MEASURE 2.5 the system is shown to be valid and reliable	[1]
Residual risk and who accepted it	The rating after the control, and the name of the person who accepted what is left.	MANAGE 1.4 negative residual risks are documented	[1] [2]
Vendor terms checked	A link to the vendor terms you read, what they say about your data, and the date you read them.	GOVERN 6.1 policies address third-party AI risks; MANAGE 3.1 third-party AI risks are regularly monitored	[1]
Last and next review	The date the row was last reviewed and when it is due again.	GOVERN 1.5 periodic review is planned, including determining its frequency	[1]

Subcategory wording is quoted or closely paraphrased from NIST AI 100-1 and the AI RMF Playbook. The choice of columns is CIO Support guidance, not a NIST template or requirement.

Six worked example rows

These rows show the level of detail that is enough. They describe a hypothetical company of about fifteen people that uses AI tools rather than building its own. The likelihood and impact ratings are illustrations on the three-point scale described below, not measurements, and your own ratings will differ with your data, customers and contracts.

Six example AI risk register rows for a small business (illustrative ratings)
Use case	Data touched and who is affected	What could go wrong (NIST AI 600-1 risk)	Likelihood x impact	Control and response	Source
ChatGPT to draft customer emails and proposals	Internal: client names, scope and pricing. Affects customers and the company.	Client details pasted into a personal account the company does not control (data privacy). A confidently wrong statement reaches a customer (confabulation).	3 x 2 = 6, High	Mitigate: company-controlled business plan only, signed in with company accounts; no Confidential or Restricted data; a person reads every draft before it is sent.	[4]
Microsoft 365 Copilot answering questions over SharePoint and OneDrive	Confidential: HR, finance and client files. Affects employees and clients.	Copilot surfaces a file to someone who can technically open it but should not see it, because sharing is wider than intended (data privacy, information security).	2 x 3 = 6, High	Mitigate: before rollout, review site permissions and organization-wide sharing links, restrict access to HR and finance files and label them, start with a small pilot group. Microsoft says Copilot 'only surfaces organizational data to which individual users have at least view permissions.'	[4] [5]
AI note-taker recording and summarizing client calls	Confidential: what clients say on calls, possibly personal data. Affects clients and call participants.	A call is recorded without the notice or consent a participant or the law expects; transcripts sit with the note-taker's vendor longer than expected (data privacy, value chain); the summary gets an action item wrong (confabulation).	2 x 3 = 6, High	Mitigate: one approved note-taker under company accounts; announce it and ask for consent at the start of each call; set retention; the call owner checks the summary before sending it; not used on calls about Restricted data. Recording and consent rules vary by state and country; take legal advice for calls across jurisdictions.	[4]
AI features a vendor switched on in the CRM or accounting software	Confidential: customer records and financials. Affects customers and the company.	A software update enables an AI feature nobody reviewed, so customer data is processed under terms nobody has read (value chain and component integration).	2 x 2 = 4, Medium	Mitigate: check AI settings after each vendor release note; read the vendor's AI terms and record the date; keep the feature off until the owner has reviewed it.	[4] [1]
AI ranking or screening of job applicants in a hiring tool	Restricted: applicant personal data. Affects job applicants.	Qualified applicants are ranked down or screened out unfairly, and the company cannot explain why (harmful bias or homogenization).	2 x 3 = 6, High	Avoid: AI does not reject or rank out applicants; a person reviews every application; take employment-law advice before any AI is used in hiring decisions.	[4]
AI chatbot answering customer questions on the website	Public content plus whatever customers type in. Affects customers.	The bot gives a wrong answer about prices, policies or warranties (confabulation), or a user's prompt makes it ignore its instructions (information security, prompt injection).	2 x 2 = 4, Medium	Mitigate: answer only from approved content; tell users they are talking to AI; hand orders, refunds and complaints to a person; review a sample of conversations each month.	[4]

Example rows and ratings are CIO Support illustrations for a hypothetical company, not data from any client and not a NIST list. Risk names in brackets are from NIST AI 600-1. The quoted Copilot sentence is from Microsoft's Data, Privacy, and Security for Microsoft 365 Copilot page, checked October 19, 2026. No other vendor's terms are described on this page.

How to rate likelihood and impact

NIST leaves the scale to you. The AI RMF says it can be used to prioritize risk but does not prescribe risk tolerance, and the Playbook's suggested action for MAP 5.1 is to establish assessment scales for measuring AI systems' impact, which may be qualitative, such as red-amber-green, and to apply them uniformly across the organization's AI portfolio. The important part is the last phrase: use the same scale on every row, or the ratings cannot be compared.

The free register uses a three-point scale for each. Likelihood: 1 means it would take an unusual mistake, 2 means it could happen in normal use, and 3 means it is likely to happen without a control, for example because the tool is used daily by several people. Impact: 1 means a minor inconvenience fixed internally, 2 means a customer is affected, money is lost or work has to be redone, and 3 means personal or regulated data is exposed, a hiring or other decision about a person is affected, or a contract or law is breached. The register multiplies the two. A score of 1 or 2 is Low (green), 3 or 4 is Medium (amber) and 6 or 9 is High (red).

Those thresholds are our suggestion for a small company, not NIST's. Change the definitions on the Scales tab if your business, customers or regulators call for something different, and record the change. Rate each row first without the control, then again with it; the second rating is your record of the negative residual risk MANAGE 1.4 asks to be documented, and someone with the authority to accept it should sign it off by name. A typed name in this spreadsheet records who accepted the risk; it is not a controlled electronic signature, and regulated quality systems have their own approval rules.

How the register maps to Govern, Map, Measure and Manage

The AI RMF Core has four functions. NIST describes Govern as cross-cutting, informing and infused throughout the other three; Map establishes the context to frame risks; Measure analyzes, assesses, benchmarks and monitors them; and Manage allocates risk resources to the mapped and measured risks on a regular basis. A small company's register touches all four. The table shows which columns carry which function, so the register can serve as evidence when someone asks how you follow the framework.

Where each NIST AI RMF function shows up in the register
AI RMF function	Register columns and tabs	Example subcategories	Source
Govern	Owner, Residual risk accepted by, Vendor terms checked, Last and next review, and the Review log tab.	GOVERN 1.5 periodic review; GOVERN 1.6 inventory; GOVERN 2.1 roles; GOVERN 6.1 third-party risk	[1] [2]
Map	Use case, Data touched, Who is affected, Risk, and Likelihood and impact.	MAP 1.1 context of use; MAP 4.1 third-party components; MAP 5.1 likelihood and magnitude	[1] [2]
Measure	Test or check done, its date and what could not be measured.	MEASURE 1.1 start with the most significant risks and document what cannot be measured; MEASURE 2.10 privacy risk; MEASURE 3.1 track risks over time	[1] [2]
Manage	Control and response, Residual risk, Status, and the Review log tab.	MANAGE 1.2 prioritize; MANAGE 1.3 respond; MANAGE 1.4 residual risk; MANAGE 4.3 incidents and errors are communicated and tracked	[1] [2]

Function descriptions are from NIST AI 100-1, section 5. The column-to-function mapping is CIO Support's.

Fill it in during one working session

A first version can usually be done in one working session of an hour or two with the owner, someone who knows the systems, and a manager from each team. Start from your approved-tools list if you have one, but expect to find uses that are not on it.

List the uses. Ask each team which AI tools and AI features they use and for what, including browser extensions, phone apps, meeting note-takers and AI switched on inside software you already pay for. Write one row per use.
Name an owner for each row. One person per use, with the authority to change or stop it. NIST's GOVERN 2.3 outcome puts responsibility for AI risk decisions with executive leadership, so in a small company the owner of a high-rated row is often the founder or CEO.
Record data and people. Write the most sensitive data the use touches and who could be hurt by a wrong or leaked output. Rows that touch Confidential or Restricted data, money, hiring or customers go to the top.
Rate it. Score likelihood and impact on the Scales tab definitions, without controls first. Keep the reasoning to a sentence in the notes.
Choose a response and a control. For each High and Medium row, pick mitigate, transfer, avoid or accept, write the control in plain words, then re-rate the row with the control in place and have the owner accept what is left.
Check the vendor terms. For each tool, read the vendor's current terms on training use, retention and where data is stored, and paste the link and the date you read them. Do not rely on someone else's summary; vendors change their terms. If a tool will touch protected health information, a HIPAA covered entity or business associate generally needs a business associate agreement with the vendor before any is entered (45 CFR 164.502(e)); reading the terms is not a substitute.
Set the review date. Review every row at least quarterly, and add or update a row whenever someone wants a new tool, a vendor switches on a new AI feature, or something goes wrong.

What the free Excel register contains

The workbook opens on a How to use tab that explains each column in one line. The Register tab has the columns above, drop-down lists for data level, response and status, a score and a Low, Medium or High rating that the sheet works out from your likelihood and impact, a next-review date that defaults to 90 days after the last review and turns red when it is overdue, and the six example rows from this page ready to overwrite.

A Scales tab holds the likelihood and impact definitions and the rating thresholds, so you can change them in one place. A NIST mapping tab lists each column with the AI RMF subcategory it supports and a link to NIST AI 100-1 and the Playbook. A Review log tab records each quarterly review, who did it and what changed.

The register pairs with CIO Support's other free AI governance downloads: the AI acceptable-use policy template and its approved-tools register, and the one-page NIST AI RMF worksheet. Use the worksheet if you want everything on one printed page, and the register when you have more than a handful of uses or want the scoring done for you. All are general guidance, not legal advice, and none is published or endorsed by NIST.

When a spreadsheet is no longer enough

A register is a good fit while AI is a set of tools your staff use. Move to a fuller program when you build AI into a product you sell, when AI makes or materially shapes decisions about people such as hiring, lending or patient care (AI that informs clinical decisions may be regulated as a medical device; take regulatory advice), or when regulated data is involved. At that point work through more of the framework's subcategories, the Generative AI Profile's suggested actions and the Playbook. If an AI use touches protected health information (PHI), CUI or GxP records, record it in the compliance program you already run, such as your HIPAA risk analysis, your CMMC scope or your GxP quality system, from the first day of use. This register does not replace a HIPAA risk analysis or the computerized-system validation a GxP quality system expects.

NIST's AI Risk Management Framework page states that AI RMF 1.0 is being revised as part of the White House AI Action Plan, and the AI RMF Playbook page says the Playbook will be updated after the revision. The subcategory numbers on this page are from AI RMF 1.0. When the revision is published, check that the numbers in your NIST mapping tab still match before you show the register to anyone as evidence.

Common follow-up questions

Does NIST publish an AI risk register template?

Not for AI. NIST AI 100-1 describes outcomes, such as inventorying AI systems and documenting the likelihood and magnitude of impacts, and the AI RMF Playbook suggests actions for each. Neither prescribes a register format. NIST does publish a general cybersecurity risk register template with NIST IR 8286 Rev. 1, which is not AI-specific. CIO Support's free register is one way to record those outcomes; it is not published or endorsed by NIST.

Is the AI risk register template free?

Yes. Download the Excel register on this page with no form, email or sign-up. It is general guidance based on NIST AI 100-1 and the AI RMF Playbook, not legal advice.

What is the difference between an AI risk register and an AI inventory?

An inventory lists the AI tools and systems you have. A risk register goes further for each use: what could go wrong, how likely and how bad it would be, what you are doing about it, and who accepted what is left. You can keep both in one spreadsheet, with the register columns added to the inventory.

How do we score likelihood and impact?

Use one scale for every row. NIST does not prescribe one; the AI RMF Playbook says scales may be qualitative, such as red-amber-green. The free register uses 1 to 3 for each, multiplies them, and rates 1 to 2 as Low, 3 to 4 as Medium and 6 to 9 as High. Change the definitions if they do not suit your business.

How often should we review the register?

NIST's GOVERN 1.5 outcome asks you to decide the frequency of periodic review. We suggest quarterly for most small companies, plus an update whenever someone adopts a new AI tool, a vendor switches on a new AI feature, or an AI incident happens.

Who should own the AI risk register?

One named person, usually the founder, CEO or an operations lead, keeps it current and runs the quarterly review. Each row also has its own owner. NIST's GOVERN 2.3 outcome puts responsibility for decisions about AI risks with executive leadership, so High residual risks should be accepted by a leader, not by the person who wants to use the tool.

Need help with this?

LAN Service Group helps small and mid-size businesses build and keep an AI risk register aligned to the NIST AI RMF, from the first inventory session and vendor-terms review to controls for high-rated uses, as a fractional CIO or as their managed IT provider, fully outsourced or co-managed alongside in-house IT staff.

Talk to LAN Service Group LAN Service Group AI services → (888) 281-7243