You can put the NIST AI Risk Management Framework into practice in about twelve weeks with a small team by working through its four functions, Govern, Map, Measure and Manage, in a deliberate order. The framework is voluntary and is not a checklist, so this calendar is one sensible sequence, not a NIST requirement.
What does the framework actually ask for?
NIST released AI RMF 1.0 (NIST AI 100-1) on January 26, 2023, and describes it as voluntary. Its core has four functions. Govern covers policies, accountability and culture. Map establishes context and identifies risks for each AI system. Measure tests and tracks those risks. Manage prioritizes and responds to them. The document says its actions do not constitute a checklist and are not an ordered set of steps, which is why you need your own plan.
NIST also published a Generative AI Profile (NIST AI 600-1) in July 2024, and its site says AI RMF 1.0 is being revised, so check for updates before you finish.
Weeks 1 to 2: inventory and ownership
Build the AI system inventory (the framework's GOVERN 1.6 expects a mechanism for this). Include models you build, models you call through an API, and AI features inside tools your staff already use. For each, record owner, purpose, data used, users affected and vendor. Name one accountable executive and one working owner per system (GOVERN 2.1 covers roles and lines of communication).
Deliverable: inventory spreadsheet and a short roles table.
Weeks 3 to 4: policy and risk tolerance
Write a short AI risk policy that states which legal and regulatory requirements apply (GOVERN 1.1), what risk level needs which review, and what is out of bounds. Decide your risk tolerance in plain terms: which harms you will not accept, and which you accept with controls (MAP 1.5 asks for risk tolerances to be determined).
Deliverable: approved policy, tolerance statement, review frequency (GOVERN 1.5 asks you to define the frequency of periodic review).
Weeks 5 to 7: map each system
For every system in scope, write a context sheet: intended purpose, deployment setting, who is affected, known limits, and how it could fail. This is the Map function. Involve people outside engineering, such as legal, support and the product owner, because the framework values interdisciplinary input (MAP 1.2). Rank systems by potential impact and take the top few into measurement.
Deliverable: one context sheet per system and a ranked risk list.
Weeks 8 to 10: measure
Choose metrics and tests for each ranked risk. Typical choices are accuracy on a held out set, bias checks across groups relevant to your use, resilience against odd inputs, and security testing of the surrounding system (MEASURE 2.7 addresses AI security and resilience). Record test sets, tools and results so another person could repeat them. Where you cannot measure something, write down why and how you will watch for it instead.
Deliverable: test plan, results log and an open issues list.
Weeks 11 to 12: manage and set the target
Decide what to do with each finding: mitigate, transfer, accept or stop the system (MANAGE 1.1 starts with deciding whether the system should proceed). Put a post deployment monitoring plan in place (MANAGE 4.1), including how incidents and errors are communicated (MANAGE 4.3). Then write a Current Profile and a Target Profile. The framework describes profiles as a way to compare where you are with where you want to be and to reveal gaps.
Deliverable: treatment decisions, monitoring plan, profile comparison and a next quarter backlog.
What to do
- Appoint an accountable owner and a small working group this week.
- Finish the AI inventory before writing any policy.
- Approve a short policy and a written risk tolerance by week four.
- Complete a context sheet for every system, ranked by impact.
- Test the top risks and keep results someone else can rerun.
- Decide a treatment for each finding and set up monitoring.
- Write Current and Target Profiles and schedule the next review.
Quick check
Is the NIST AI RMF a certification?
No. It is a voluntary framework, and NIST does not certify organizations against it.
Do I need all four functions for every system?
Apply them in proportion to risk. Low impact systems can get a lighter pass than the systems you ranked highest.
How Safeguard helps automate this
The slow parts of this calendar are the inventory and the evidence. Safeguard's compliance suite covers frameworks, controls, evidence, continuous monitoring, automated tests and questionnaires, so evidence is gathered from connected systems and code instead of by screenshot. Technical evidence comes from your real code and dependencies, including scans and SBOMs, which shows which AI libraries and model packages you actually ship. Monitoring continues between reviews, so the team can spend its time on measurement. Autonomous remediation can open tested fix pull requests, but it is off by default, each category starts on Manual, and you pick which categories run automatically. Safeguard helps prepare evidence; it does not certify anything. Book a call.
Sources
Last reviewed: 18 October 2026.
