Trained in March, hired in July
Workforce AI training is the only evidence in your binder that expires without anyone touching it, because its validity depends on a user population that turns over, so it has to be attached to access provisioning rather than delivered once as a go-live event.
Pull the completion roster for the training that went with your busiest AI tool. Then pull the list of people who used that tool last week, and count the names on the second list that are not on the first.
Nobody did anything wrong. The roster is accurate, the training happened, and the number is not zero.
Every other artifact in your governance binder is a statement about the organization. A committee charter, a disclosure policy, a model card, a register row: each one stays true until somebody changes it, and when it goes wrong you know, because you were in the room when it went wrong.
A training record is a statement about people. It is a join between a curriculum and a population, taken on one date. The curriculum sits still. The population does not.
It moves through hiring, transfers, the float pool, agency and locum cover, the academic changeover that puts a new cohort in front of the same screens every July, and the ordinary spread of a tool into a second department because somebody in the first one liked it.
So the record stays literally true and quietly stops being an answer. “Did we train the workforce on this tool?” is answered by the roster. “Are the people using this tool trained on it?” is answered by the join. Those are the same question only on go-live day.
That is the part worth naming. This evidence decays without anyone touching it. Nothing turns red, and the artifact looks exactly as good as it did the week it was filed, which is why it is usually the last thing re-examined and the first thing to fail a side-by-side check.
The frameworks treat training as a standing capability rather than a project deliverable. The NIST AI Risk Management Framework, published by NIST in January 2023, places workforce competency inside GOVERN, alongside the policies and accountabilities expected to persist. The Joint Commission’s September 2025 guidance with the Coalition for Health AI names staff education in the same structural position. Neither describes a class you hold once. The control mapping sits in the reference library.
The durable version attaches training to the only process that already moves at the speed of your workforce: access. Role-based provisioning has a request, an approver and a date on it already. When the entitlement that switches the tool on is the thing that requires the completion, the two lists cannot drift apart, because the same event creates both.
Monday
Send one email to whoever administers access for your highest-volume AI tool and ask for two things in the same reply: the current active-user list, and the completion roster for the training that accompanied go-live. Put them side by side. The gap is not a compliance failure, it is a clock reading — it tells you how many months of turnover that record has absorbed since the day it was true.
The rest of your binder waits for you to break it. This one expires on the payroll’s schedule, and nothing in the building will tell you when.
One operational argument a week
The Operations Edge lands each Monday: a hook, one thing to use before lunch, and the full argument here in the archive. No vendor sponsorship, ever.
The instruments behind the writing
Every framework in the series is published as a working file: registers, protocols, audit rubrics and unit-economics models, sized to be used rather than admired.
See the toolkits The library← The override rate nobody has ever pulled
Published under the Institute's editorial standard.
Author: Neel Chauhan, MD MBA, physician-executive and founder of the Healthcare AI Institute. Last reviewed against the standard on 2026-09-01.
Drafted as issue 11 of The Operations Edge, argued from operating experience rather than from a dataset. No turnover figure is used because no dated, named source was available for one; the two framework references are cited by publisher and date.
The Institute accepts no vendor sponsorship, holds no vendor equity and takes no referral fees.