Safety & Governance

How to Turn AI Agent Risks into a NIST AI RMF Register in Excel

10/23/2026

AI-agent risks are easier to review when each entry describes a concrete event, its possible impact, and who is responsible for following up. This guide shows how to turn those entries into a filterable Excel register mapped to relevant parts of the NIST AI Risk Management Framework (AI RMF).

The workbook is a practical tracking tool, not a NIST-issued template or a universal scoring system. You’ll record evidence, responses, ownership, and review dates, then check that the register can support ongoing review as the agent or its operating context changes.

Quick answer

Start with one row per risk and fields for the agent context, scenario and impact, relevant NIST function and subcategory, assessment evidence, owner, response, status, and due or review dates. Use the NIST AI RMF Playbook to find relevant references and prompts, but select only what fits your system and use case.

For agent-specific starting points, consider untrusted content influencing tool use, model or data integrity problems, and unintended actions. Define any likelihood, severity, or combined rating scales locally; they are workbook conventions, not NIST-prescribed scores.

In Excel, turn the register range into a table with Ctrl+T so you can filter entries and extend the register more easily.

Step-by-step guide

Prepare the agent context and workbook environment

Before opening Excel, define the agent’s intended use and the environment it operates in. Note the tools it can call, the permissions it has, what systems or people could be affected, and where its inputs come from; these details help make later risk entries specific to this agent rather than generic labels.

Use desktop Excel to create the workbook, and consult the NIST AI RMF Playbook in Excel or another available format. The Playbook offers suggested actions and references aligned to the framework’s four functions, but treat them as adaptable prompts, not a required checklist or an agent-specific register template.

NIST’s framework supports ongoing risk management across the AI lifecycle, so plan to revisit the register as the agent or its context changes.

Sources: Artificial Intelligence Risk Management Framework (AI RMF 1.0); NIST AI RMF Playbook; CAISI Issues Request for Information About Securing AI Agent Systems; More on data validation.

Define the register fields around decisions and follow-up

Create a header row in Excel with fields for Risk ID, agent and use context, risk scenario, potential impact, NIST function, NIST subcategory, assessment evidence, likelihood, severity, local rating, owner, response, status, due date, and next review date.

Use one row per risk, and make the scenario and impact separate columns so reviewers can see what could happen and why it matters. Record the evidence used for the assessment, not just a rating.

Define any likelihood, severity, or combined-rating scales in the workbook; these are local conventions, not NIST-prescribed scores. Map relevant functions and subcategories to entries, but don’t treat the functions as a required sequence: GOVERN is cross-cutting.

Sources: Artificial Intelligence Risk Management Framework (AI RMF 1.0); NIST AI RMF Playbook; CAISI Issues Request for Information About Securing AI Agent Systems; More on data validation.

Write agent-specific risk scenarios and map them to the framework

In the Risk scenario and Potential impact fields, describe what could happen in this agent’s context and why it matters. For example, untrusted retrieved content could influence a tool call, a model or data integrity issue could affect the agent’s output, or the agent could take an unintended action while pursuing its objective.

For each row, consult the NIST AI RMF Playbook and select function and subcategory references that fit the scenario and use case. Record the references in the columns you created, and keep only relevant prompts; the Playbook is a guide, not a mandatory checklist.

The mapping helps reviewers trace each risk to the framework without implying that the functions must be followed in sequence.

Sources: Artificial Intelligence Risk Management Framework (AI RMF 1.0); NIST AI RMF Playbook; CAISI Issues Request for Information About Securing AI Agent Systems; More on data validation.

Record assessment, ownership, and response

For each risk row, record the evidence behind your assessment—for example, a test result, incident record, or review—so another person can understand the basis for the entry. If you use likelihood or severity ratings, define what each scale means in the workbook; NIST does not prescribe a universal scoring formula.

Assign an accountable person or team, then record the chosen response, current status, due date, and next review date. Check that the owner can act on the response and that open work has a clear follow-up date.

Keep the NIST function and subcategory mapping relevant to the scenario; the functions are not a required sequence, and GOVERN is cross-cutting. Revisit the assessment as the agent or its context changes.

Sources: Artificial Intelligence Risk Management Framework (AI RMF 1.0); NIST AI RMF Playbook; CAISI Issues Request for Information About Securing AI Agent Systems; More on data validation.

Create a filterable Excel table

Enter the headers and initial risk rows you prepared, keeping one risk per row. Select the full range, press Ctrl+T, and confirm that the table has headers.

Give the table a recognizable name, such as RiskRegister, so it’s easier to identify as the workbook grows. Use the filter controls in the header row to review entries by owner, status, NIST mapping, or date; add a test row and confirm it appears within the table and can be filtered.

Sources: Artificial Intelligence Risk Management Framework (AI RMF 1.0); NIST AI RMF Playbook; CAISI Issues Request for Information About Securing AI Agent Systems; Using structured references with Excel tables; More on data validation.

Optionally flag overdue open items

If an automatic flag would help with follow-up, add an Alert column to the RiskRegister table and enter this formula in its first data row: =IF(AND([@[Due date]]<TODAY(),[@Status]<>"Closed"),"OVERDUE",""). Excel should fill the calculated column down the table; the formula marks dates before today as overdue unless the status is Closed.

Test it with an open row due before today and a closed row with the same date: only the open row should show OVERDUE. Also check a future due date, which should remain blank.

This is an optional workbook aid, not a NIST requirement; confirm the formula works in the Excel version and locale where the register will be maintained.

Maintain the register through review

Set the next review date for each open risk, then revisit the entry on that date and when the agent’s tools, permissions, inputs, or operating context changes. Use new test results, incidents, or other assessment evidence to update the scenario, assessment, response, status, and owner; add a clear follow-up date for any work that remains open.

The register should evolve with the agent, not become a one-time checklist.

To check that follow-up is practical, filter the table by review date, owner, and status. Confirm that each item due for review has an accountable owner and a recorded next action, and update the row after the review.

Check that entries are usable and traceable

Use the table filters to review entries by owner, status, NIST function or subcategory, and due or review date. Check that each row describes a specific agent event and potential impact, cites assessment evidence, identifies an accountable owner, and records a response and a date for follow-up.

Confirm that each framework mapping fits the scenario; the AI RMF functions are not a required sequence, and GOVERN is cross-cutting. If you added the optional Alert formula, compare its output with an overdue open row, an overdue closed row, and a future-dated row.

The open overdue item should be flagged, while the others should remain unflagged; test the formula in the Excel version and locale used for the workbook.

Resolve Excel setup issues

If Data Validation is disabled while you’re adding optional dropdowns or other consistency checks, check whether the worksheet is protected or the workbook is shared. These conditions can prevent changes to validation settings; use an editable desktop workbook to make the changes, then test them in the version your team uses.

If you need validation in Excel for the web, set it up first in desktop Excel and test the shared workbook afterward. If the optional overdue formula returns an unexpected result, check that its column headers and status values match the formula, then test it in the intended Excel version and locale; date and formula behavior can vary across environments.

Next steps

Before relying on the workbook, filter it by owner, status, NIST mapping, and date. Check that each entry gives a specific scenario and impact, links to assessment evidence, names an owner, and has a response and follow-up date.

If you use a local scoring scale or an optional overdue alert, document the scale and test the formula in the Excel version and locale your team will use. Then set a review cadence and revisit entries when the agent’s tools, permissions, inputs, or context change; the register is most useful when it remains part of ongoing risk management.