Example scenario walkthrough
What a stress test finds before your council sees the draft.
A policy that reads well can still fail the first time someone has to apply it. Between the first and second draft, every AI adoption policy we write is run against written case scenarios. This is what that produced on one municipal engagement.
How the test is run
The documents are used, not read.
Each scenario is written as though the policy and its procedures had already been adopted and were in force, then walked through the way a CAO or privacy officer would actually use them: which provision is engaged, in what order, who must act, what happens next, and what the documents do not say. Every clause relied on is quoted before it is relied on. Where a draft leans on another instrument (the privacy policy, the acceptable-use policy, an HR policy), that instrument is read too, so a cross-reference is never taken on trust.
Three rules govern the exercise. No problem is manufactured: where the documents handle a situation, the result says so. Every fix is written as replacement or additional text that keeps the existing clause numbers, so nothing has to be renumbered. And where a gap sits outside what a municipal policy can fix (a provincial sanction regime that does not yet exist, for example), the result says that plainly rather than proposing a clause that would overreach.
Results at a glance
Seven scenarios, and what each one found.
Every scenario is hypothetical and described by role. Clause numbers, instrument names and the municipality itself are generalized here; in the working paper each one is cited exactly.
| # | Scenario | What held | What needed attention | Verdict |
|---|---|---|---|---|
| 01 | A council member enters closed-session material into a personal AI tool | What heldThe prohibition is unambiguous and the reporting route works. The drafts also decline to invent a councillor sanction the province has not created. | What needed attentionThe suspension power has nothing to act on when a personal account was used. The documents should say what the entry means: a release of information beyond the municipality’s control. | Handled, fix proposed |
| 02 | A front-counter employee pastes a ratepayer’s tax complaint into a free AI tool | What heldTwo independent prohibitions catch it, the interim permission gives no cover, and the breach response and discipline routes are complete. | What needed attentionTraining and acknowledgment only trigger on being issued an approved tool, so the employee most likely to breach is the one never trained. Directors’ duties do not name incident reporting. | Handled, fix proposed |
| 03 | The CAO uses an approved tool on a director’s performance material | What heldHigh-risk activity correctly identified, and the report goes to the head of council rather than to the CAO: the best-drafted clause in either document. | What needed attentionAfter the report is received, the documents fall silent on who reviews, decides and applies a consequence. The breach response still runs through an officer who reports to the CAO. | Decision needed |
| 04 | A development permit is refused with AI assistance; the applicant asks whether a machine decided | What heldThe verification record and the “no final decision by AI” rule together show that a person decided, and show the working. A strong position to be in. | What needed attentionRead literally, the high-risk rule would require privacy-officer approval on every permit file, which no department can operate. Nothing directs anyone to fix the collection notices on the municipality’s forms. | Decision needed |
| 05A | A contracted engineering firm uses its own AI tool on municipal road data | What heldScope and consequence provisions reach the contractor correctly. | What needed attentionThe approved-tool appendix is the wrong instrument for a contractor’s internal tooling. The contract should govern it, and the drafts do not say so. | Handled, fix proposed |
| 05B | The IT provider switches on an AI feature inside licensed help-desk software | What heldThe integrated-AI rule is exactly on point, and the breach is identified and routed correctly. | What needed attentionNothing requires the IT provider to ask first, nothing names who orders it off, and the procurement gate does not open on a vendor-initiated change. | Handled, fix proposed |
| 06 | AI note-taking and transcription of council meetings | What heldThe public portion has a usable, lawful path: standing approval from the chair, notice, minutes remain the record, verification before release. Personal note-takers are prohibited outright. | What needed attentionThe procedural bylaw was read and is silent on recording, so nothing conflicts. The in-camera position is a council choice, and it should be made knowingly rather than inherited from a template. | Decision needed |
The pattern was consistent.
The drafts were strong at saying what is and is not permitted. Every prohibition, every characterisation and every reporting route landed correctly, and in three places they handled a hard case better than they had to.
Where they were thinner is in saying who acts next: after a report reaches the head of council, when a vendor feature must be switched off, when a collection notice on a form is wrong, when a high-risk activity recurs daily. That is the shape of the thing a stress test finds and a careful read does not.
One scenario, drawn out
A front-counter employee pastes a ratepayer’s letter into a free AI tool.
Hypothetical. Roles only; no person described is a real person. Follow it role by role, or across the five stages: what happened, who is told, what the privacy officer does, who is notified, and what follows. Each box names the instrument that authorizes the step. Solid boxes are what the first drafts already do; dashed boxes are the two fixes that went into the second draft.
| Role | 1 · What happened | 2 · Who is told | 3 · Containment | 4 · Assessment | 5 · Notify, follow through |
|---|---|---|---|---|---|
| Front-counter employeeon a municipal workstation | 1 · What happenedDay 1: the eventPastes a ratepayer’s tax complaint (name, roll number, address, a description of financial hardship) into a free web AI tool on a municipal workstation, using a personal login created that morning, to draft a reply. | 2 · Who is toldRules engagedAI procedure: not an approved AI service (the approved list is empty) · interim permission excludes an identifiable ratepayer · a personal AI account · high-sensitivity information (financial) without approval · AI policy: approved services only | 3 · ContainmentWorkstation activity may be examinedAcceptable-use policy: no expectation of privacy, breaches investigated; privacy policy logs show what was pasted | 5 · Notify, follow throughProposed: training reaches everyoneIn the first draft, training and sign-off trigger only when someone is given an approved tool. This employee never was, so was never trained and never signed. Proposed: every employee, within 30 days. | |
| Directorof the employee’s department | 1 · What happenedDay 2: hears about itThe employee mentions it, pleased with the result. The director now knows of a suspected breach. | 2 · Who is toldReports immediatelyTo the privacy officer or the CAO: AI policy; AI procedure; privacy and access policy (breach reporting); HR policy | 3 · ContainmentProposed: name the dutyThe directors’ row lists “gaps or new AI-related risks”, not incidents. The duty exists; the row people read should name it. | ||
| Privacy and access officerdelegated under the privacy and access policy | 2 · Who is toldOpens the breach filePrivacy and access policy: breach file and form. A breach is any unauthorized disclosure of municipal information, personal or not. | 3 · ContainmentContainsAsks the provider to delete the upload and chat history; preserves the workstation logs, under the privacy and access policy (containment) | 4 · AssessmentAssesses the levelConfidential-class information to an external recipient. Real-risk-of-harm factors: POPA s. 10(2); Alta Reg 143/2025 s. 4 | 5 · Notify, follow throughDecides the notificationsThe privacy officer decides, not the director, not the employee. Then documents findings and reports to the CAO. | |
| Outside the municipalityAI provider · ratepayer · commissioner | 3 · ContainmentAI providerDeletion requested. No agreement binds it, so nothing can be compelled: the whole problem. | 4 · AssessmentRatepayer notifiedPrivacy and access policy; notice contents POPA s. 10(2)(a), Reg s. 4(3) | 5 · Notify, follow throughCommissioner notifiedPrivacy and access policy; POPA s. 10(2)(b), Reg s. 4(4) | ||
| CAOhead under the access and privacy statutes | 1 · What happenedNothing to suspendSuspension under the AI policy reaches approved services; the employee had none. Acceptable-use rules apply instead. | 2 · Who is toldInformedAI procedure; AI policy. No conflict, so the council route is not engaged. | 4 · AssessmentSteers the follow-throughReceives the findings; discipline runs on its own track; considers whether training or the approved-services list needs to change. | ||
| Management and human resourcesseparate track from the breach file | 3 · ContainmentDiscipline: separately from the breach fileThe privacy policy passes discipline to management and HR. AI policy: corrective or disciplinary action up to termination, the standard the acceptable-use and HR policies already set. Mitigation the employee will raise: never trained, never signed. | 5 · Notify, follow throughThen: training and sign-offAI awareness training and the acknowledgment, under the AI procedure | |||
Verdict: handled, fix proposed.
Every rule, characterisation and reporting route lands, and nobody has to invent a process. Nothing here turns on where the tool’s servers are: the failure is the disclosure to a service the municipality does not control (no agreement, no safeguards, no authority to disclose), and the privacy policy’s breach process is built for exactly that.
The two dashed boxes are the gaps. The training and acknowledgment gates miss the very employee this scenario describes, and the directors’ row does not name the reporting duty. Both were wording changes for the second draft, not changes of rule.
Where this sits
Between the first draft and the second.
The stress test is not a separate product. It runs between phase four and phase five of the municipal method, after the two documents are drafted and before administration and council review them, which is the last point at which changing the wording costs nothing.
Drawn from live work for an Alberta municipal district and published with the scenarios, clause references and instrument names generalized. Every situation is hypothetical; no person or event described is real.
Municipalities across Canada
Have a draft that has never been used?
Whether we wrote it or you did, a policy is worth applying to a handful of real situations before council adopts it. That conversation is free and it takes about fifteen minutes.
Book a 15-minute call