NIST's AI Risk Management Framework, Minus the Jargon

Most AI governance advice is written for companies with a compliance department. You have twelve people, one of whom does the books on Thursdays. So when someone tells you to adopt the NIST AI Risk Management Framework, the reasonable reaction is to close the tab.

Don't. The framework is shorter and more flexible than its reputation. Strip the vocabulary out and it's four questions you can answer in an afternoon, plus a habit you keep up afterward.

What the framework actually says

NIST released AI RMF 1.0 on January 26, 2023. It's voluntary and free. As of August 2026 it's still version 1.0, though NIST notes on that page that the framework "is being revised as part of the White House AI Action Plan," so expect movement.

The core is four functions: GOVERN, MAP, MEASURE and MANAGE. Per the AI RMF Core, GOVERN is the cross-cutting one that sets the culture and the paperwork. MAP builds "the context to frame risks related to an AI system." MEASURE applies "quantitative, qualitative, or mixed-method tools, techniques, and methodologies to analyze, assess, benchmark, and monitor AI risk." MANAGE allocates "risk resources to mapped and measured risks on a regular basis."

Here's the line small companies never get told about. The same document says "some organizations may choose to select from among the categories and subcategories; others may choose and have the capacity to apply all categories and subcategories." NIST expects you to skip things. A twelve person firm is exactly who that sentence was written for.

GOVERN: one named owner, one page

For a small business, GOVERN collapses to three decisions. Who owns AI risk management here by name. What are you not willing to let an AI system do without a person signing off. Where does that get written down so a new hire can read it.

That's it. One page in your shared drive beats a policy binder nobody opens. Put the owner's name at the top, list the tools you've approved, list what's off limits (client financial records, health information, anything you'd hate to see quoted back to you), and date it.

MAP: inventory before analysis

MAP is the step people skip, and it's the one that pays. You cannot manage risk in a system you haven't written down. Walk your actual workflow and answer these for each place AI touches it:

• What does this system do, and what decision does its output feed?
• What data goes in, including anything a customer typed or a staff member pasted?
• Who gets hurt if the output is confidently wrong, and how would you find out?
• Is there a human between the output and the customer, or not?
• If the vendor changed the model tomorrow without telling you, what breaks?
• Can you turn it off in five minutes?

Most Metro Vancouver businesses we talk to are surprised by the inventory itself. The AI receptionist is on the list, sure. So is the sales rep pasting client emails into a free chatbot, and that one is usually the bigger exposure.

NIST also names seven characteristics of trustworthy AI: valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. Useful as a checklist. But note the framework's own honesty about the catch: "tradeoffs are usually involved," including "between predictive accuracy and interpretability." You don't get all seven maxed out. You decide which ones matter for this use and say why.

MEASURE: pick numbers you'd actually notice

Measurement at your scale is not a benchmark suite. It's three or four numbers reviewed monthly, and one of them should be a quality sample.

If you run an AI phone agent, that's booking rate, escalation rate, and twenty randomly pulled transcripts a month read by a human. If it's an email drafter, it's how often a draft goes out unedited versus rewritten. The rewrite rate is the honest signal. Nobody's marketing page reports it.

The generative-specific failure modes are catalogued in NIST's companion Generative AI Profile (NIST AI 600-1), published July 26, 2024. NIST's announcement describes it as identifying 12 risks and just over 200 actions, including "a lowered barrier to entry for cybersecurity attacks" and systems "confabulating or 'hallucinating' output." You won't do 200 actions. Read the 12 risk names and mark the three that apply to you.

MANAGE: decide the response before you need it

Manage is the fire drill. Who takes the call when a customer says the AI told them something wrong. What's the rollback, and has anyone tested it. What gets logged so you can reconstruct what happened.

Write those three answers down next to your one page policy. Then actually schedule the monthly review, because a framework you run once is just a document.

The Canadian layer NIST doesn't cover

NIST is American and says nothing about your privacy obligations. In BC you're likely under PIPEDA, BC's Personal Information Protection Act, or both.

Federal, provincial and territorial privacy authorities published principles for responsible, trustworthy and privacy-protective generative AI on December 7, 2023. The one that matters most for a small deployment: "accountability for decisions rests with the organization, not with any kind of automated system." If your AI quotes a wrong price, that's your price.

This isn't theoretical anymore. On May 6, 2026 the federal Privacy Commissioner, together with Quebec's CAI and the BC and Alberta commissioners, released findings in a joint investigation of OpenAI concluding that the way OpenAI initially collected personal information to train GPT-3.5 and GPT-4 "was overbroad and therefore inappropriate." Your obligations as a user of these tools aren't identical to a developer's. But BC's regulator is clearly engaged, and "it was on the internet already" is not a defence.

Where this doesn't apply

Some honest limits, because the pitch usually leaves these out.

If you have no AI in production yet, doing the full framework first is procrastination with a citation. Do the one page and the inventory. Skip the rest until something is live.

The measurement guidance is genuinely weak for tiny deployments. NIST asks for repeatable, objective testing. With forty calls a month you don't have the sample size for that, and anyone who tells you otherwise is selling a dashboard. Read transcripts instead and accept that it's qualitative.

And adoption itself is less universal than the noise suggests. Statistics Canada reported that in the second quarter of 2026, 19.2% of Canadian businesses used AI to produce goods or deliver services in the previous 12 months, up from 12.2% a year earlier and 6.1% in Q2 2024. Real growth, and still four in five businesses not doing it. Construction sat at 9.2%. If your trade is in that band, waiting a year is a defensible business decision, not a failure of nerve.

One more from that release, and it cuts against the small business reader. Cybersecurity or privacy concerns were a barrier for 11.6% of businesses with 1 to 4 employees, compared with 30.0% of those with 100 or more. Small firms report less concern. That gap probably reflects less scrutiny rather than less risk.

Sources

• National Institute of Standards and Technology, "AI Risk Management Framework," 2023 (updated). https://www.nist.gov/itl/ai-risk-management-framework
• NIST Trustworthy and Responsible AI Resource Center, "AI RMF Core," AI RMF 1.0, 2023. https://airc.nist.gov/airmf-resources/airmf/5-sec-core/
• NIST Trustworthy and Responsible AI Resource Center, "Characteristics of Trustworthy AI Systems," AI RMF 1.0, 2023. https://airc.nist.gov/airmf-resources/airmf/3-sec-characteristics/
• NIST, "Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile" (NIST AI 600-1), July 26, 2024. https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence
• NIST, "Department of Commerce Announces New Guidance, Tools 270 Days Following President Biden's Executive Order on AI," July 2024. https://www.nist.gov/news-events/news/2024/07/department-commerce-announces-new-guidance-tools-270-days-following
• Statistics Canada, "Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026," June 11, 2026. https://www150.statcan.gc.ca/n1/pub/11-621-m/11-621-m2026010-eng.htm
• Office of the Privacy Commissioner of Canada and provincial and territorial privacy authorities, "Principles for responsible, trustworthy and privacy-protective generative AI technologies," December 7, 2023. https://www.priv.gc.ca/en/privacy-topics/technology/artificial-intelligence/gd_principles_ai/
• Office of the Privacy Commissioner of Canada, "PIPEDA Findings #2026-002: Joint Investigation of OpenAI OpCo, LLC," May 6, 2026. https://www.priv.gc.ca/en/opc-actions-and-decisions/investigations/investigations-into-businesses/2026/pipeda-2026-002/

If you'd rather not read a federal framework on a Sunday, we'll do the translation with you. Autana Solutions is based in Burnaby and builds AI employees for small teams across the Lower Mainland, governance page included, not bolted on later. Book a free call and we'll go through your inventory in half an hour.