Yes, a small team can run the NIST AI Risk Management Framework from a spreadsheet. Build tabs for an AI inventory, governance, mapped context, measures and a risk register, then review them on a schedule. Automate once the inventory changes faster than people can update it or evidence must come from live systems.
What is the AI RMF, and is it certifiable?
NIST released AI RMF 1.0 on 26 January 2023 for voluntary use. It is organised around four functions: Govern, Map, Measure and Manage. NIST publishes a companion Playbook with suggested actions, and a Generative AI Profile (NIST AI 600-1) released on 26 July 2024. NIST's page also states that AI RMF 1.0 is being revised as part of the White House AI Action Plan, so check the current version before you start.
There is no certificate attached to the AI RMF itself. That changes the goal: you are building a defensible record that you understand and manage your AI risks, for customers, boards and regulators who ask.
How should the workbook be laid out?
Seven tabs cover the four functions without inventing paperwork.
1. AI inventory. One row per system or model use: name, owner, purpose, vendor or in house, model and version, data used, users affected, deployment date, status. Include shadow use such as staff using a chat assistant for customer work.
2. Govern. Policies and roles. Columns: policy, owner, approval date, next review, link. Add an accountability row for who can approve a new AI use and who can stop one.
3. Map. Context per system: intended use, known limits, who could be harmed, legal and contractual constraints, third party components, impact severity. This is where you decide which systems need deeper treatment.
4. Measure. For each system, what you track: accuracy or error rates on your own test set, bias checks where relevant, security tests, monitoring for drift, incident count. Columns: metric, method, threshold, last result, date, tester.
5. Manage: risk register. Risk, system, likelihood, impact, treatment, owner, due date, status. Link rows back to Map and Measure.
6. Manage: incidents and changes. Dated log of model changes, prompt or data changes, and any failures, with response and lessons.
7. Evidence index. One row per artefact with file location, date, and which function it supports.
Keep the tabs linked by a system ID so a filter shows everything about one system.
Who does the work, and how often?
- Inventory: system owners update when they ship, and a coordinator reconciles monthly against procurement and engineering.
- Govern: leadership reviews annually, or when AI use changes materially.
- Map: done before launch and re-done on major changes.
- Measure: scheduled per system. High impact systems get more frequent tests than internal drafting aids.
- Manage: risk review quarterly. Incident log is continuous.
Name a single workbook owner. Most spreadsheet programs fail from drift, and drift is an ownership problem before it is a tooling problem.
When does a spreadsheet stop working?
Look for these signals.
- The inventory is wrong on the day you audit it, because teams add models faster than rows.
- Measure results live in notebooks and dashboards, and copying them in is manual and late.
- More than one person edits at the same time and versions diverge.
- Customers send questionnaires that map your workbook to other frameworks, and each answer takes hours.
- Software supply chain facts matter, such as which model libraries and dependencies each system uses and whether they carry known vulnerabilities.
If two or three of these are true, move the parts that change fastest, the inventory and the evidence, onto tooling that collects them from the source. Keep the human decisions, such as impact ratings and risk acceptance, with named people.
Which mistakes hurt most?
- Treating the AI RMF as a policy exercise and never measuring anything.
- Leaving out third party AI features embedded in products you buy.
- No thresholds, so a measure can never fail.
- Unowned risks that sit open for months.
- Copying the Playbook into cells wholesale. Use it as a menu of ideas and pick actions that fit your systems.
What to do
- Read the current AI RMF and Playbook pages and decide which Generative AI Profile risks apply.
- Build the seven tabs and assign an owner to each.
- Fill the inventory from procurement, engineering and a short staff survey.
- Map and rate impact for each system, then deep dive on the top ones.
- Define metrics with thresholds and run the first round of tests.
- Review the risk register in a quarterly meeting with minutes.
- Revisit this workbook when NIST publishes the revision.
Quick check
Is the NIST AI RMF mandatory?
NIST describes it as voluntary. Other rules, such as contracts or sector regulation, may require risk management practices that the AI RMF helps you evidence.
Can one workbook cover several AI systems?
Yes. Use a system ID on every tab and filter by it. Split into separate workbooks when ownership or confidentiality requires it.
How Safeguard helps automate this
Safeguard's compliance suite covers frameworks, controls, evidence, continuous monitoring, automated tests and questionnaires, so evidence flows from connected systems and code rather than screenshots, and monitoring keeps running between reviews. For AI systems that rely on open source model libraries and packages, scans, SBOMs and reachability give you factual input for the Map and Measure tabs. Safeguard's agent can open tested fix pull requests for vulnerable dependencies and other findings. It is off by default, every category starts on Manual, and you choose which categories run automatically and the approval level. Safeguard helps prepare and keep evidence ready. Impact ratings and risk acceptance stay with your people.
Open the framework mapping demo or book a call.
Sources
Fetched 18 October 2026.
Last reviewed: 18 October 2026.
