The definition

What is shadow AI?

Shadow AI is the use of artificial intelligence tools (ChatGPT, Microsoft Copilot, Google Gemini, free AI writing assistants, AI transcription services) by employees without authorization from their organization. It is called "shadow" because it operates outside official IT governance, procurement reviews, and policy frameworks. The organization doesn't know it's happening, which means it can't protect against the risks it creates.

The term is an evolution of "shadow IT", the long-standing problem of staff using unsanctioned software. AI tools have extended the pattern because many are free, capable, and genuinely useful for exactly the kinds of tasks municipal staff do every day: drafting documents, summarizing meetings, answering routine questions, generating form letters.

48% of public servants surveyed by KPMG Canada use AI tools for work. 39% say they do not have, or are not aware of, a clearly communicated data policy. (KPMG Canada, 2025; 349 respondents from federal, provincial and municipal governments.) Nearly a third of 282 local and regional governments surveyed by MNP in 2025 have no formal policies or guidelines for AI. In our view, the gap between usage and governance is widest in small and rural municipalities, where policy capacity is most limited.

What's actually happening

What municipal staff are using AI to do

Shadow AI in municipalities is not dramatic. It doesn't look like a breach. It looks like a staff member getting their work done faster. Common scenarios include:

  • Pasting a ratepayer complaint into ChatGPT to draft a response letter, including the ratepayer's name, address, and the nature of their complaint.
  • Uploading audio from a council meeting to a free AI transcription tool to generate meeting minutes, including discussions that may reference individuals by name.
  • Using an AI writing assistant to draft a bylaw or permit correspondence that includes civic addresses and property owner details.
  • Asking an AI tool to summarize a performance review or HR document to prepare for a staff meeting.
  • Using a consumer AI tool on a personal device, while working from home, to process work that involves citizen data.

In each of these scenarios, personal information (information that is subject to provincial privacy legislation) has been entered into a system that your municipality did not procure, has no data processing agreement with, and has no visibility into. Whether that data was used to train the AI model, retained by the platform, or simply processed and discarded, your municipality may have no way of knowing. And without a policy, you have no agreed way to respond if something goes wrong.

The legal exposure

Why shadow AI is a privacy problem, not just an IT problem

Canadian municipalities are public bodies subject to provincial privacy legislation. In Alberta, that means the Access to Information Act (ATIA) and the Protection of Privacy Act (POPA). Comparable legislation governs public bodies in other provinces, such as MFIPPA in Ontario, FOIPPA in BC and FIPPA in Manitoba.

These laws are explicit: personal information may only be collected, used, and disclosed for purposes authorized under the legislation. When a staff member enters citizen data into an external AI platform without authorization, that can amount to a disclosure: the personal information has been shared with a third party outside the municipal system. Depending on the circumstances and the data involved, this may be a privacy breach that may need to be reported to the provincial privacy oversight body and affected individuals. Reporting rules differ by province and are generally tied to the risk of significant harm; in Ontario, mandatory breach reporting to the IPC starts January 1, 2027.

The absence of malicious intent doesn't change the legal exposure. A staff member who uses ChatGPT to draft a better response to a difficult ratepayer has not done anything wrong from a personal standpoint; they were trying to do their job well. But the municipality that has no policy governing that behaviour, no training explaining the risk, and no Privacy Officer accountable for the outcome is in a fundamentally different position than one that does.

The response

How to address shadow AI without banning it

A prohibition alone rarely works. Staff are likely to keep using tools they find helpful, whether or not a policy exists, especially when those tools genuinely make their work better and the prohibition feels arbitrary. The effective response governs AI use rather than attempting to eliminate it.

A compliant response has five components:

  • Designate a Privacy Officer: someone formally accountable for AI governance, with the authority to make decisions about sanctioned tools and the obligation to manage breaches.
  • Conduct a personal information inventory: understand what data your staff are working with so you can assess the risk that shadow AI creates.
  • Draft and adopt an AI Acceptable Use Policy: specify which tools are sanctioned, what data may and may not be entered, and the consequences of unauthorized use. This is the governance control.
  • Deliver staff training: explain both the risk and the approved alternatives. Staff who understand why the policy exists are more likely to follow it.
  • Build a review cycle: the AI landscape changes faster than any policy written today will remain current. Quarterly review of sanctioned tools is not excessive given the rate of change.

This is not a technology problem. It is a change management problem, which is why the most important first step is not finding the right software but designating a human being who is accountable for it.

Common questions

Frequently asked questions

Is shadow AI common in Canadian municipalities?

It is common in Canadian public bodies. KPMG Canada (2025) found that 48% of public servants surveyed across federal, provincial and municipal governments use AI tools for work, and 39% say they do not have, or are not aware of, a clearly communicated data policy. In our view, the gap is often wider in small and rural municipalities, where staff use AI tools because they are genuinely helpful, with no awareness that doing so may violate provincial privacy legislation.

Does Microsoft Copilot count as shadow AI if we already use Microsoft 365?

It depends on how staff access it. Copilot Chat is included at no extra cost for eligible Microsoft 365 users signed in with a work (Entra) account, with enterprise data protection; the full Microsoft 365 Copilot licence costs extra. If staff are accessing Copilot through personal Microsoft accounts or using the free web version (copilot.microsoft.com) without organizational provisioning, that use is unsanctioned and counts as shadow AI. Even for the work version, we recommend reviewing Microsoft's data protection terms before sanctioning it.

What if our staff are using AI on personal devices?

The device ownership does not change the analysis. If an employee enters municipal data (ratepayer information, council materials, HR records) into an AI tool on a personal device, the privacy exposure is the same. Your AI Acceptable Use Policy must address both work and personal devices when used for work purposes.

What are the consequences if we ignore shadow AI?

The most immediate risk is a privacy breach that damages public trust and may need to be reported to the provincial privacy oversight body and affected individuals. In Alberta, the longer-term risk is being found non-compliant with the Privacy Management Program requirement under POPA, which has been in effect since June 11, 2026. Municipalities with no shadow AI governance and no Privacy Management Program are in a weak position if a complaint is filed.