The four functions, in one line each
GovernThe culture, policies and accountability. NIST treats it as cutting across the other three, which is why the pack starts here rather than with tools.
MapEstablishing context: what AI is in use, for what purpose, with what data, and what could go wrong.
MeasureAssessing and tracking what was mapped, using methods that produce a record rather than an opinion.
ManageActing on what was measured: prioritizing, responding, overriding, recovering and reviewing.
Control by control
| Control | Primary | NIST subcategories | How the artefact answers it |
|---|---|---|---|
| 01 AI Tool Check | Map | MAP 1.1, GOVERN 1.6 | Documents every AI tool actually in use, its purpose, the data it touches and who introduced it. NIST asks for an inventory mechanism; the AI Tool Check is the first pass and the inventory is what it feeds. |
| 02 AI Tool Inventory | Govern | GOVERN 1.6, GOVERN 1.7, MAP 1.1 | The standing record: approved tools, an owner for each, data classification, and the decision to retire one. Covers the decommissioning subcategory most small firms never document. |
| 03 One-Page AI Usage Policy | Govern | GOVERN 1.2, GOVERN 3.2, GOVERN 4.1 | States permitted and prohibited use, who decides grey areas, and the required human review. This is where trustworthiness characteristics become an actual instruction to staff. |
| 04 Traffic-Light Risk Matrix | Measure | MAP 5.1, MEASURE 1.1, MANAGE 1.2 | Rates each use green, amber or red with the reason and the reviewer, so likelihood and impact are recorded and treatment is prioritized on evidence rather than instinct. |
| 05 Prompt Registry | Measure | MEASURE 2.9, MEASURE 3.1, MAP 2.2 | Captures reused prompts that touch a client or feed a decision, with the output reviewed and the known limits noted. This is the small-firm substitute for formal model documentation. |
| 06 Vendor Vetting Checklist | Govern | GOVERN 6.1, GOVERN 6.2, MAP 4.1, MANAGE 3.1 | Answers the third-party questions before data moves: training use, retention, location, IP terms, and what happens when the vendor fails. Monitored again at each quarterly review. |
| 07 One Hour Staff AI Training | Govern | GOVERN 2.2, MAP 3.4, MAP 3.5 | Delivers the training subcategory outright, and it is the only control that produces attendance, a knowledge check and a completion record. Practitioner proficiency and human oversight, evidenced. |
| 08 Agentic AI Kill-Switch Protocol | Manage | MANAGE 2.4, MANAGE 2.3, GOVERN 1.7 | Names what an autonomous agent may never do, who can stop it, how, and inside what time. NIST's deactivation subcategory reads as though it was written for this control. |
| 09 Quarterly AI Governance Review | Manage | GOVERN 1.5, MANAGE 4.1, MANAGE 4.3, MEASURE 3.3 | The 90-minute cycle: what changed, what failed, what is overdue, incidents logged and communicated, next date set. Without it a pack is a document rather than a management system. |
| Opportunity Register (bonus) | Map | MAP 1.4, MAP 3.1, MANAGE 1.2 | Records the business value of each proposed use and its state from proposed to implemented, so the upside is governed on the same page as the risk. |
Where the pack does not reach, and what to say
| Measurement depth | MEASURE 2.1 to 2.13 assume test sets, benchmarks and TEVV personnel. A ten-person firm has none of that. The pack substitutes documented human review and the prompt registry, and says so. |
| Independent review | MEASURE 1.3 asks for internal experts who were not front-line developers. In a small business the adviser is that second pair of eyes, which is worth naming in the engagement letter. |
| Impact assessment | MAP 5.2 expects structured engagement with affected communities. Out of scope for an SME, and pretending otherwise would be the fastest way to lose credibility. |
How to use this in the room
Do not open with the framework. Open with the client's own shadow AI list, then use this page to answer "is this a real standard or something you made up". Govern, Map, Measure, Manage is the whole speech. Then say the line that wins the engagement: this is first line documentation aligned to NIST AI RMF 1.0, it is not a certification, and nobody can certify you against it.