TL;DR: Govern sets policies, accountability, risk tolerance and culture, and applies across everything. Map establishes each system's context, purpose, stakeholders and risks. Measure assesses those risks with tests, metrics and evaluation. Manage prioritises and treats risks, handles incidents and monitors over time. In practice: Govern produces an AI policy and owners; Map produces a register entry and threat model; Measure produces evaluation and red team results; Manage produces release decisions, risk acceptances and incident playbooks.
How to approach it
Name each function, give its intent in one line, then the concrete artefacts it produces. Mention the Generative AI Profile and the framework's voluntary status.
A strong answer
Map
context, purpose, risks
Measure
tests, metrics, evals
Manage
treat, respond, monitor
Govern
Policies, accountability, risk tolerance, culture
| Function | Intent | Artefacts |
|---|---|---|
| Govern | Culture, policy, roles, accountability, third-party oversight | AI policy; RACI per system; review board; training; vendor requirements; risk tolerance statements |
| Map | Understand context and identify risks | Use-case register entry; intended use and misuse; data sources; affected people; threat model; regulatory classification |
| Measure | Analyse, assess and track risks | Evaluation suites; security testing and red teaming; bias and robustness tests; monitoring metrics; documented methods and limits |
| Manage | Prioritise and act on risks | Mitigations; release gates; risk acceptances with owners and dates; incident response; decommissioning plans |
How they interact. Govern underpins the rest. Map, Measure and Manage form a loop: what you measure depends on what you mapped, and what you manage feeds back into the map when systems or contexts change.
Generative AI. NIST AI 600-1 (July 2024), the Generative AI Profile, lists risks new to or amplified by generative AI, including information security risks such as prompt injection and data poisoning, and suggests actions organised by the four functions.
Status. The RMF is voluntary and not certifiable. As of November 2026 it remains version 1.0, with NIST revising it following the July 2025 US AI Action Plan.
Security engineer's angle. Measure is where AI security testing lives; Manage is where findings become controls and release decisions; Map is where threat models belong. Tying your security work to these functions makes it legible to governance teams and auditors.
What interviewers probe next
- "Where does a red team fit?" Measure, with results flowing into Manage and back to Map.
- "Who should own Govern?" Senior leadership with a cross-functional body; security, legal, privacy, product and engineering all participate.
- "How do you avoid paperwork?" Make each subcategory point to an artefact the engineering process already produces.
What if?
What if · interview follow-ups
Hold your answer as the constraints change
A new team member asks what the four functions of the NIST AI Risk Management Framework are, and how they show up in the team's daily work on an AI assistant.
Start here
What does Govern cover?
Who decides what risk is acceptable?
Policies, accountability, risk tolerance and culture, applying across all the other functions. In practice it produces an AI policy, named owners for each system, a risk appetite statement and training. Without it, the other functions have no one accountable for their outputs.
Carry it forwardGovern sets policy, ownership and tolerance.
Weak answerTreating Govern as a one-time policy document.
Constraint changes
What does Map produce?
What do you need to know about a system before testing it?
Each system's context, purpose, stakeholders and risks: a register entry and a threat model. For the assistant, that means who uses it, what data and tools it touches, who could be harmed and which risks apply.
Carry it forwardMap establishes context and risks per system.
Weak answerSkipping Map and testing generic risks.
Constraint changes
What does Measure produce?
How do you know a mapped risk is under control?
Assessments of the mapped risks with tests, metrics and evaluation: evaluation suites, red-team results, bias and privacy assessments, with rates and conditions. Measure should test the risks Map identified, not a generic list.
Carry it forwardMeasure tests the risks Map found.
Weak answerMeasuring what is easy rather than what was mapped.
Constraint changes
What does Manage produce?
What happens after the evaluation results arrive?
Decisions and actions: prioritising and treating risks, release decisions and risk acceptances, incident playbooks and ongoing monitoring. It closes the loop by feeding incidents and monitoring back into Map and Measure.
Carry it forwardManage turns results into decisions and responses.
Weak answerEnding the process at the evaluation report.
Constraint changes
What if the team treats the functions as sequential phases?
What happens when the system changes after launch?
They are not a waterfall. Govern runs throughout, and Map, Measure and Manage repeat whenever the system, its use or the threat changes. A new tool added to the assistant should reopen Map, prompt new Measure work and a fresh Manage decision.
Carry it forwardThe functions repeat across the lifecycle.
Weak answerCompleting the RMF once at launch.
Now defend it
How would you explain the RMF in practice?
Govern produces policy, owners and risk tolerance across everything; Map produces register entries and threat models per system; Measure produces evaluations and red-team results for the mapped risks; Manage produces release decisions, risk acceptances, incident playbooks and monitoring. The functions repeat on every material change, not once.
How it's scored
Weak: Names the functions only.
Hire: Explains each function's intent and gives artefacts.
Strong Hire: Shows the loop and Govern's cross-cutting role, connects security work to functions, cites the GenAI Profile and current status.
Common mistakes
- Treating the functions as one-time sequential steps.
- Claiming certification.
- Measurement without action in Manage.