{"resourceId":"wellington-north-ai-policy-copilot-no-decisions-20261014","versions":[{"version":"external-1cd8c3d76f447f70eb100fb802461fa023cc1e6e1f641b35edf08da1010594d4","resource":{"id":"wellington-north-ai-policy-copilot-no-decisions-20261014","title":"Township of Wellington North, Ontario approves AI policy allowing Copilot for drafting while barring AI from permits, hiring and other decisions","organization":"Township of Wellington North, Ontario","sector":"Small local government staff productivity and policy","geography":"Wellington North, Ontario, Canada","publishedAt":"October 14, 2026","publicationDate":"2026-10-14","eventDate":"2026-10-05","sourceName":"Wellington Advertiser","sourceLabel":"Independent local news report","sourceUrl":"https://www.wellingtonadvertiser.com/wellington-north-approves-artificial-intelligence-use-for-government-work/","evidenceClass":"independent-reporting","outcomeClass":"emerging","topics":["governance-procurement","knowledge-work","operating-model"],"finding":"Council approved a township AI policy at its Oct 5 meeting (reported Oct 14). Microsoft Copilot is the approved tool; public models may be used only with already-public information. Copilot may draft correspondence, reports, policies and presentations, summarize materials and assist research, but users must independently verify facts, figures, dates, calculations and quotations. The policy prohibits using Copilot to make decisions on spending, approving permits, hiring, discipline or staff evaluations, and bars relying on AI summaries for recommendations, decisions, approvals, voting or actions. \"Responsibility for township business decisions cannot be delegated to an AI tool.\" Policy covers employees, council, committees, volunteers, contractors and consultants; FOI/privacy compliance required; annual review. CAO noted AI may be inaccurate, incomplete, biased or outdated.","sledRelevance":"Small-government template transferable to U.S. towns/counties: allow productivity copilots while hard-excluding permitting and HR decisions. Canadian FOI/privacy context differs; decision boundary is the transferable lesson.","evidence":"Wellington Advertiser article opened in full via WebFetch and curl. Quotes CAO Brooke Lambert and policy language; township policy PDF not separately opened.","architectureImplications":"Approved-tool preference for Copilot in M365 tenancy; public-model use restricted to public information implies DLP/guidance rather than hard technical blocks alone.","governanceImplications":"Explicit non-delegation of decisions; annual review; covers contractors/consultants as well as staff and council.","securityPrivacyImplications":"Must comply with freedom of information and privacy laws; public models only when inputs are already public.","caveats":"Secondary report; no measured productivity results; enforcement mechanisms beyond policy language not described.","streamIds":["local-government"],"roles":{"sales":"Sales framing: Wellington North is a small-township pattern—approve Copilot for drafting, forbid AI on permits and HR decisions. Sell policy-aligned M365 Copilot enablement with decision-exclusion training, not autonomous permitting. Discovery: which decisions must stay human, what is already public vs confidential and how FOI will treat prompts. Bounded engagement: acceptable-use policy, Copilot pilot for report drafting and an annual review metric set. Interpretation: transferable to U.S. towns; do not claim Canadian privacy rules map 1:1.","engineering":"Engineering framing: Fit is Microsoft 365 tenants enabling Copilot with DLP for non-public data and guidance blocking public-model use on confidential inputs. Prerequisites: licensing, sensitivity labels and a written decision-exclusion list (permits, hiring, spending). Proof of value: sampled draft quality after human verification and zero AI-originated permit/HR decisions in audit. Interpretation: controls on data classification matter more than model choice.","delivery":"Delivery framing: Roll out Copilot to defined roles, train verification and decision bans, and schedule annual policy review. CAO owns policy; IT owns tenancy controls. Acceptance: staff attest to decision exclusions; random audit of AI-assisted reports shows human verification. Risks: shadow use of public ChatGPT with non-public data and over-trust of summaries before votes. Interpretation: keep permitting and HR systems out of AI write paths."},"roleDetails":{"sales":{"customerProblem":"Small township wants staff productivity from AI without delegating permits, hiring or spending decisions.","stakeholders":["CAO","Council","Staff authors","IT","Contractors/consultants"],"discoveryQuestions":["Which systems issue permits or HR actions?","What data is public vs confidential?","How will FOI treat Copilot prompts?","What annual review metrics will council want?"],"valueHypothesis":"Copilot may speed drafting if humans verify; no measured time savings reported.","potentialEngagement":"Copilot enablement plus decision-exclusion training and a 90-day drafting pilot with verification audit.","unsupportedClaims":["Measured productivity gains","That public ChatGPT is safe for any township data","AI can approve permits under this policy"]},"engineering":{"fit":"Small local governments on Microsoft 365 seeking governed staff copilots.","architectureIntegration":"Copilot in tenant with sensitivity labels; block or guide public-model use for non-public inputs; no write-back to permitting/HR decision systems.","prerequisites":["M365 Copilot licensing","Data classification","Written decision-exclusion list"],"constraints":["Canadian FOI/privacy","Annual policy review","Contractors in scope"],"security":"Public models only with already-public information; confirm DLP and logging.","proofOfValue":["Sampled draft accuracy after human edit","Audit finding zero AI permit/HR decisions","Staff attestation completion"]},"delivery":{"work":"Enable Copilot, train verification and exclusions, audit samples, annual review.","dependencies":["Licensing","CAO policy ownership","IT tenancy admin"],"ownership":"CAO office with IT support.","skillsAdoption":"Staff learn to verify facts/figures/dates and never accept AI summaries as decision basis.","governanceCheckpoints":["Policy adoption (done)","90-day pilot review","Annual policy review"],"acceptanceCriteria":["Proposed: verification checklist used on AI-assisted council reports","Proposed: no AI write access to permit/HR decision workflows","Proposed: annual review completed on schedule"],"risks":["Shadow public-model use","Over-reliance on summaries before votes","Contractor noncompliance"]}},"retrievedAt":"2026-10-15T22:22:30-05:00","enrichedAt":"2026-10-15T22:28:00-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Positions AI as assistive with mandatory human verification; councillors stressed checking and rechecking public information.","procurementImplications":"Names Copilot as approved tool without disclosing license counts or cost in the article.","operatingModelImplications":"Drafting/summarization allowed; permitting, spending, hiring and evaluation decisions remain human-only.","sourceVerification":{"openedUrl":"https://www.wellingtonadvertiser.com/wellington-north-approves-artificial-intelligence-use-for-government-work/","referenceExcerpt":"Responsibility for township business decisions cannot be delegated to an AI tool.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}