Leadership · lesson 4 of 4 · 14 min
Lead responsible change
Create an adoption program that combines literacy, participation, measurement, and a real ability to stop what is not working.
- Set measures beyond tool adoption
- Build participation into the design
- Establish the authority to stop
Literacy is now an obligation, not a perk
Training has moved from good practice to stated requirement. The EU AI Act obliges providers and deployers to ensure a sufficient level of AI literacy among staff operating AI systems. ISO/IEC 42001 requires demonstrated competence for people whose work affects AI management system performance. NIST's AI RMF treats workforce capability as a governance function. Whatever the jurisdiction, the direction is one way: you will be asked to evidence that your people were trained, and when.
A requirement that an organization ensure staff involved in operating or overseeing AI systems have sufficient knowledge to do so responsibly, and can evidence that this was provided.
A 900-person agency set four measures: percentage of high-tier systems with evaluation evidence less than 90 days old, registered uses per quarter, near-misses reported, and proposals stopped. In year one they stopped two, and registrations tripled. Both numbers were reported as successes, which is what made the third and fourth measures credible.
Measure outcomes, not seat time
Licenses issued and hours completed measure spend, not value. Measure the things that indicate the program is working: cycle time on the specific workflows you redesigned, quality scores against the fixed evaluation sets, the number of AI uses registered (a rising number is a healthy sign, not a problem), incidents and near-misses reported, the proportion of high-tier systems with current evaluation evidence, and how many proposals were stopped. A program that has never said no is not governing.
The people doing the work know where it breaks
Adoption designed in a leadership offsite fails at contact with the work. The staff running the process know which step is judgment and which is drudgery, and they will tell you if telling you is safe. Make it safe: no headcount implications tied to the pilot, a real channel for reporting that something is producing bad output, and visible instances of that feedback changing the plan.
Reserve the right to stop
The capability that distinguishes a governance program from a compliance exercise is the ability to end something. Name in advance who can stop a deployment, what evidence triggers it, and what happens to the work in the meantime. Then use it once, publicly, on something that is not working. Nothing else you do will communicate as clearly that the controls are real.
Draft the one-page program
Write the first page of an AI adoption program: four measures, one participation mechanism, and the stop authority.
One page. If it needs more, the parts that do not fit are not yet decisions.
0 of 5 practice steps checked. Saved on this device.
Measures: high-tier systems with evaluation evidence under 90 days old; registered uses per quarter; near-misses reported; proposals stopped. Participation: a monthly 30-minute practitioner review, chaired by the process owner, not by IT. Stop authority: the CISO, on any unclosed high-tier finding or any unacceptable-failure recurrence. Literacy evidence: completion records and curriculum version, retained three years. Cadence: quarterly to the risk committee.
Check your understanding
Answer both to mark the lesson complete. You can try again. This is practice, not a graded exam.
1. Which is the strongest indicator that an AI governance program is working?
2. Why does the EU AI Act matter to an organization outside the EU?
Three things to remember
- Workforce AI literacy is becoming an evidenced obligation, not a nice-to-have.
- Measure outcomes: evaluation currency, registrations, near-misses, and things stopped.
- Name the stop authority and use it once, visibly.

The full oversight framework behind this path, across identity, data, AI, vendors, and risk.
Examples are composites written for teaching unless a source is cited. Figures come from the sources listed where given.