Every member of staff takes a twenty-minute scored exercise on AI in practice. The leadership team spends an hour on where the group split. The organisation keeps this: a two-page record for its policy file, executive summary first, detail second. The organisation below is fictional; the shape is what you would hold.
This is the record a head of learning and development, or a head of compliance, puts in front of the chief executive. It is written to be read in two minutes: the numbers, five lines of findings, two charts and the actions with owners and dates on page one; the five decisions, the policy lines and the notes on data and scoring on page two. Example Organisation does not exist. Its people, dates and figures are illustrative. The five decisions are the ones in Module 1 of AI for Professional Practice. The public version is open to anyone with no sign-in, if you want to take it before reading on.
Executive summary. Prepared 20 November 2026 for the Risk Committee of 3 December 2026.
Share of the 58 who chose the option the answer key marks unsafe. Terracotta where more than four in ten did; those two decisions are where the leadership hour went.
Share in each team who chose the unsafe option, with the count. Darker is higher. The support team handles the most internal documents and is where line 4 and the tool list matter most. Leadership figures are four people.
| What | Owner | Date | Status |
|---|---|---|---|
| Exercise opened to all 70 staff, 9 to 16 November (58 completed); leadership hour held on the anonymised group results | Head of L&D | 19 Nov 2026 | Done |
| Approved AI tool list issued to all staff | Head of IT | 26 Nov 2026 | Open |
| Twelve outstanding completions, including two leadership team members (L&D tracks) | Line managers; Chief Executive for the two | 30 Nov 2026 | Open |
| Five policy lines to the Risk Committee for approval | Chief Operating Officer | 3 Dec 2026 | Scheduled |
| Exercise re-run for all staff; the decision 4 (redacted extract) result reported to the committee | Head of L&D | 12 Dec 2026 | Open |
| Fewer than one in five choose the unsafe option at decision 4 on the re-run | Chief Operating Officer | 12 Dec 2026 | Target |
| Decision point, and what it tests | Policy line drafted | |
|---|---|---|
| 1 | The fluent draftA confident draft arrives with a source nobody can find. Approve it, or hold it.Tests whether a draft that reads well goes out on the strength of how it reads. | No AI-assisted draft leaves the organisation until the person who signs it has checked every source it cites. |
| 2 | Who answers for the workA junior colleague ran the first draft through a tool. Whose work is it?Tests whether responsibility stays with the person whose name goes on the work. | The person who signs the work answers for it in full, whoever or whatever produced the first draft. |
| 3 | The four checksSources, quotes, figures, claims. Which to run when the work goes out tonight.Tests whether the checks that catch a fabricated source and a true figure in the wrong place are run before use. | Four checks before any AI-assisted draft is used: the source exists, the quotation matches its source, the figure is attached to the right thing, the claim holds. |
| 4 | The redacted extractAn internal document with the names taken out, and a public tool that would summarise it in seconds.Tests whether removing names is treated as making confidential or personal information safe for a public tool. | No confidential, commercially sensitive or personal information, redacted or not, goes into any tool that is not on the organisation's approved list. |
| 5 | Use, disclose or declineA tool would save a day on a client report. Use it and say so, use it without saying, or decline.Tests whether the judgement clients rely on the organisation for stays with the person, and when a client should be told. | The organisation's published statement says how AI tools are used; the responsible manager decides, case by case, what more a client needs to hear. |
Each person sees their own result: every decision as it was put to them, the choice they made, the marks it carried, and the reasoning behind the answer the key marks right. A certificate on passing, with a verification code, for as long as the record is held.
The organisation's named administrator sees the group view and each person's result for the people in their own organisation. In this organisation, line managers and HR do not see individual results. The system ranks nobody.
Each person was told what is recorded and who sees it before they started; the wording is filed with this record. LearnFrame opens a person's record only to support the organisation or to keep the service running.
Data held. For each person: the decision at each of the five points, the score, the attempt number and the date, plus the account data needed to run the service (name, work email, sign-in log). No ranking, no inference about anyone's character. Held on LearnFrame Central in the EU under a written processing agreement for 24 months from the exercise, the organisation's chosen period, then deleted or anonymised; earlier on written instruction.
Scoring. Fixed against an answer key written by LearnFrame and open to the organisation's review. No language model scores or writes any person's record.
Status. A training record and a draft of policy, not a qualification and not legal or regulatory advice. The organisation decides whether and how it appears in its own training records.
Each person signs in with a work email on LearnFrame Central and takes Module 1: five scored decisions in a branching scenario about AI-assisted work. Nothing to install. Each person sees their own result and reasoning at the end.
By video call, on the anonymised group results: where the organisation agreed, where it split, and which teams need which line. The policy lines are drafted in that hour, for the organisation's own approval.
Two pages for the policy file, in the organisation's name, with owners, dates and a target. The exercise can be run again after the policy lands, so the record shows a before and an after.
If a page like this would be useful for your organisation, arrange a conversation. The same shape works for dilemmas of your own: the scenario changes, the record does not.