Feature
There’s a particular kind of dread that appears when legal asks whether the new LLM feature is ready for release. Not in the “does the demo work” sense. In the “will this end up in a hearing” sense. That moment reveals something. The AI risk program, if it exists at all, might be running on trust and habit.
Martin Stanley, senior advisor at the National Institute of Standards and Technology (NIST) and author of the AI Risk Management Framework, said the quiet part out loud. “These systems are incredibly vulnerable when they're not protected properly, and too often the risk program runs on intuition rather than structure.”
Table of Contents
- NIST AI Risk Management Framework: At a Glance
- How NIST Translates Risk Into Action
- What 'Trustworthy AI' Actually Requires
- Why the AI Risk Profile Is the System’s Memory
- The First 90 Days of a Real AI Risk Program
- The Due Diligence Most Vendors Hope You Skip
- The Most Useful Work Is Often Invisible
NIST AI Risk Management Framework: At a Glance
The NIST AI Risk Management Framework doesn’t solve the problem for you. What it does is force the conversation to move out of Slack and into a system. One that surfaces questions at the right time, with the right people in the room and with a structure for how the answers get recorded. Once that happens, risk stops being ambient and starts being real.
The framework operates through four verbs:
- Govern
- Map
- Measure
- Manage.
They sound procedural. They aren’t. They create a rhythm that connects accountability to action, context to evidence, failure to remediation. And they do this without handwaving.
Related Article: AI Risk Management: How to Secure GenAI, Agentic AI and Shadow AI
How NIST Translates Risk Into Action
Let's break down the four parts of the NIST AI Risk Management Framework:
Govern
Govern assigns responsibility. It doesn’t ask whether your team is responsible. It asks who, when and what artifacts they maintain.
This step includes naming decision-makers, mapping their roles to the AI lifecycle and making the entire structure visible across the organization. An enterprise governance strategy turns risk from a theory into a function.
Map
Map gives the system a biography: Purpose, data sources, interfaces, stakeholders, environments, potential for misuse.
The map becomes the story of what the system is and why it exists. This story, once written, creates the conditions for trust. Because now anyone in a position of responsibility can trace the intent, the constraints and the tradeoffs. Without that, you’re not managing a system. You’re tolerating it.
Measure
Measure replaces assurance with evidence. It shifts the conversation from “we think the model performs well” to “here is how it behaves under pressure.”
The evaluation plan includes performance metrics, robustness tests, privacy stress scenarios, bias evaluations and explainability outputs. These are not checklists. They are engineering artifacts. They live with the system, versioned and visible.
Manage
Manage transforms the analysis into operations. The system doesn’t just produce outputs. It also ages. It fails. It collides with the world. Management means planning for that. It means selecting and implementing controls, handling exceptions and preparing for events. It also means defining retirement criteria. Every AI model expires — only some are retired with intention.
Stanley grounds it in practice: “Govern is where you create what we call a risk-aware culture… and then map, measure and manage turn that awareness into actual decisions. That’s how you move from theory to practice.”
What 'Trustworthy AI' Actually Requires
The framework highlights seven traits: valid, safe, secure, private, fair, transparent, explainable. Each one describes a state you can demonstrate through evidence.
- Valid means the system includes task-specific evaluations and confidence intervals.
- Safe means abuse cases have been imagined, logged and reviewed.
- Secure means it has gone through adversarial inputs and fault injection.
- Private means data minimization has been applied, with leak scenarios explored.
- Fair means subgroup performance is visible, with remediation plans prepared.
- Transparent means decisions create an audit trail.
- Explainable means the system offers a rationale that people can follow directly.
Each trait leads to the same review question: who tested this, when and what did they see?
Why the AI Risk Profile Is the System’s Memory
The profile is what holds the shape of the work: It captures purpose, constraints, operating assumptions and risk tolerance. It names the outcomes that matter most and draws a clear line around which risks the team accepts.
The strongest profiles stay specific. They call out AI risks like prompt injection, retrieval leakage, summarization bias and model drift. They name controls, isolation boundaries, rate limits, subgroup evaluations, rollback triggers. And they pair each one with evidence — screenshots, logs, numbers that show what happened and when.
When a profile becomes part of the system artifact, it moves with the code. That’s how governance reaches scale.
“The generative AI profile identifies twelve risks that are either unique to or exacerbated by these systems, and it maps each one to specific suggested actions,” said Stanley. "It’s a way to get past vague policy into actionable controls.”
The First 90 Days of a Real AI Risk Program
Every organization wants a plan. The ones that make progress usually stop at 90 days.
Form an AI risk council. Write a charter. Select two priority systems. Create profiles. Run baselines for performance, robustness, privacy and bias. Open a register. Schedule a tabletop drill. Implement a few high-leverage controls. Wire monitors to known risk signals. Publish a report. Tune metrics. Retire one control. Automate one flow of evidence. Update your profiles to match the world that exists.
What you build in that window becomes the foundation. Not because it solves everything, but because it creates a path.
The Due Diligence Most Vendors Hope You Skip
Vendor selection is a risk transaction. When a third party provides AI, your team becomes accountable for its behavior. So ask the questions you ask your own teams:
- Who owns the risk?
- What tests were run?
- How is the system monitored?
- What happened after it failed?
- Can we see the profile?
- Can we see the logs?
Webinar
Dec
17
[EIS Webinar] Beyond the Pilot: Why Most GenAI Projects Fail to Scale and How to Become One of the Success Stories
Move from experimental projects to integrated solutions that drive strategic decision-making.
Register
Webinar
On demand
From Manual to Magical: How AI Transforms CX Teams
Learn how to replace manual support processes with automation that actually delivers.
Watch Now
Webinar
On demand
How to Build a Solid Knowledge Foundation for AI Success
See how leading brands keep their AI honest, compliant and actually helpful.
Watch Now
Webinar
On demand
Fix the Content Bottleneck: Build a Better WebOps Strategy
Content stalled? Dev overloaded? You’re not the only one. Learn how streamlined WebOps bridges the publishing gap.
Watch Now
Webinar
On demand
Beyond Storage: Smarter Content, Bigger Impact with DAM + AI
Discover how the DAM + AI duo makes content smarter, stronger and more accessible.
Watch Now
Webinar
On demand
Agentic AI Playbook: Real-World Customer Service Use Cases You Can Deploy Now
Boost self-service by 30% and slash call volume by 63% with agentic AI.
Watch Now
If the answers are vague or delayed, so is the trust.
Related Article: AI Governance Isn’t Slowing You Down — It’s How You Win
The Most Useful Work Is Often Invisible
You don’t need a new team to govern AI risk. You need your current teams to speak the same language. Profiles should update when roadmaps shift, monitoring should trigger when risk levels change, exceptions should expire and logs should link to the systems they describe.
Organizations that do this well don’t talk about AI ethics in abstract terms. They write their decisions down, show them and improve them.
