Curriculum/Leadership/Lesson 1 of 4

Leadership · lesson 1 of 4 · 14 min

Govern AI by risk

Classify AI use cases by impact and assign oversight that is proportionate, practical, and accountable.

What you will be able to do

  • Build a three-tier use-case classification
  • Set oversight that scales with impact
  • Run an intake that people actually use

Learn · part 1 of 4

One policy for everything fails twice

A single AI policy applied uniformly is simultaneously too heavy for the marketing team drafting captions and too light for the eligibility system deciding who gets a benefit. The result is predictable: shadow AI everywhere, and the genuinely consequential systems reviewed with the same checklist as a slide deck. Tiering is not bureaucracy. It is what makes the heavy controls affordable where they matter.

In plain language
AI use-case register

A maintained inventory of every AI system in use, each with an owner, a risk tier, its data classification, evaluation evidence, and its decommissioning trigger.

Example

A state agency tiered 61 AI uses in six weeks. Fifty-three were low, six moderate, two high. The two high-tier systems, an eligibility screening pilot and a resume filter, absorbed most of the governance effort, and the eligibility pilot did not proceed. Under the previous uniform policy, both had been approved with the same four-question form as a chatbot for drafting newsletters.

Learn · part 2 of 4

Classify by consequence, not by technology

Tier by what happens to a person when the system is wrong. Low: internal productivity, easily reversible, no personal impact. Moderate: customer-facing content, operational decisions with a human in the loop, limited personal impact. High: decisions affecting rights, safety, money, employment, housing, credit, health, liberty, or eligibility, or any use at scale that is hard to reverse. The model's sophistication is irrelevant to the tier.

Learn · part 3 of 4

Match the control set to the tier

Low: acceptable-use policy, data tiering, a register entry. Moderate: named owner, documented evaluation, human review before external release, quarterly check. High: documented impact assessment, pre-deployment evaluation with a fixed test set, bias testing where people are affected, a monitored human override, an appeal route for the affected person, and a written trigger for shutting it off. Keep an inventory. A control you cannot point at in a register does not exist.

Learn · part 4 of 4

Make intake easier than avoidance

Governance programs fail at intake, not at policy. If registering an AI use takes three weeks and a committee, people stop registering. A one-page form, a 48-hour answer for low tier, and a named human who helps rather than blocks will surface more real risk than any policy document. Measure the intake queue like a service, because it is one.

Practice · try it yourself

Tier your first ten

List ten AI uses in your organization and assign each a tier using consequence to a person as the only test.

Include the uses that were never formally approved. Those are the ones the register is for.

List ten AI uses, including unofficial ones
For each, write what happens to a person when it is wrong
Assign low, moderate, or high on that basis alone
Name an owner for every high and moderate entry
Write the decommissioning trigger for each high entry

0 of 5 practice steps checked. Saved on this device.

Knowledge check · 2 questions

Check your understanding

Answer both to mark the lesson complete. You can try again. This is practice, not a graded exam.

1. What determines an AI use case's risk tier?

2. Registrations of AI use have collapsed since the policy launched. The most likely cause is:

Recap · take it with you

Three things to remember

  1. Tier by consequence to a person; one uniform policy fails at both ends.
  2. Match control depth to tier, and keep every use in a register with an owner.
  3. Make the compliant path faster than the workaround or the data goes dark.
Cover of The Boundaryless Enterprise

Go deeper

The Boundaryless Enterprise

The full oversight framework behind this path, across identity, data, AI, vendors, and risk.

View on Amazon ↗

Sources

  1. NIST, AI Risk Management Framework 1.0 (2023)
  2. EU AI Act, high-level summary of risk tiers
  3. ISO/IEC 42001, Artificial intelligence management system (2023)

Examples are composites written for teaching unless a source is cited. Figures come from the sources listed where given.