Municipal governments and AI: FIPPA-compliant adoption

The City of Brampton put generative AI use under an interim staff policy in February 2024, months before Ontario's Information and Privacy Commissioner issued formal guidance on the topic. Cities moved first. Regulators are still catching up, and that gap describes most of municipal AI adoption in Canada right now.

A clerk in a mid-sized Ontario city who wants to use an AI tool to draft a council report, summarize a bylaw complaint, or triage a freedom-of-information request is not looking for a definition of FIPPA. She wants to know whether she can put resident data into the tool without creating a compliance problem for her employer. The answer depends on where the data goes, who can access it, and whether the company processing it is Canadian or American.

What FIPPA Requires

Ontario's Freedom of Information and Protection of Privacy Act, and its municipal sibling MFIPPA, do not mention artificial intelligence anywhere in their text. Both were written for paper records and mainframe databases. What they regulate is the collection, use, retention, and disclosure of personal information by public institutions, and a municipality's AI vendor contract falls inside that scope the moment resident data touches the tool.

The Information and Privacy Commissioner of Ontario published guidance in late 2023 addressing generative AI, following up with a checklist for public sector institutions in 2024. The IPC did not create new law. It applied existing FIPPA principles, among them necessity, purpose limitation, and security safeguards, to a technology municipal IT departments were already piloting without much legal review. British Columbia's Office of the Information and Privacy Commissioner took a similar approach under that province's FIPPA, publishing generative AI guidance in June 2023, one of the earliest such releases by any Canadian regulator.

"Public bodies remain accountable for personal information in their custody or control, regardless of the technology used to process it," the BC OIPC wrote in its 2023 guidance. Procurement officers now cite that line back to vendors during contract review.

A municipality cannot outsource its FIPPA obligations to a vendor's terms of service. If an AI tool stores prompts on servers outside Canada, or uses submitted data to train a model available to other customers, the institution that fed it resident information still answers to the IPC.

The US Jurisdiction Problem

Most of the AI tools municipal staff already use without official sanction, consumer chatbots, browser extensions, free-tier assistants, are built by US companies subject to US law, including the CLOUD Act. That statute lets US authorities compel American providers to turn over data those providers control, regardless of where the servers physically sit. For a municipality handling property assessments, bylaw enforcement records, or social housing applications, that is the specific fact a privacy impact assessment has to address.

This is where the distinction between a US-headquartered AI vendor and a Canadian one starts to matter for procurement, not just for privacy officers filling out forms. Augure, a Toronto-based vendor, says it has no US corporate parent and no US investors, and that customer conversations, documents, and AI inference are never handled by US-jurisdiction providers. That is a scoped claim about customer content. It is not a claim that the CLOUD Act has no reach anywhere in the world.

The scoping matters, and municipalities doing due diligence should notice the difference between it and a blanket promise that data never leaves Canada. Augure's own privacy documentation discloses that inference for certain model tiers runs on EU infrastructure under zero-retention agreements, used for some tiers by default and for failover on others, and that payment processing and email delivery still involve US-based service providers, as is true of nearly every SaaS vendor selling into Canada. None of that is unusual. What is unusual, among AI vendors courting the municipal market, is disclosing it plainly in a privacy policy rather than making a claim that collapses under a procurement officer's first follow-up question.

Procurement Officers Ask Different Questions

Municipal IT procurement has historically evaluated software on uptime, cost, and integration with existing systems. AI tools added a fourth column: where the model provider sits jurisdictionally, and what happens to the data after the prompt is submitted.

Procurement staff at Ontario and BC municipalities say they now ask vendors, according to conversations reported at municipal IT conferences in 2024 and 2025, a consistent set of questions. Does customer data train the vendor's models, and can that be disabled by contract? Where does data sit at rest, versus where does inference happen? Does the vendor have a US corporate parent or US institutional investors? What is the retention period, and is there a hard deletion mechanism? Has the vendor mapped its sub-processors, including for email and payment processing?

Those questions did not appear in municipal RFPs three years ago. They appear now because IPC guidance, combined with a handful of embarrassing incidents involving staff feeding confidential records into public chatbots, forced the issue onto the desks of clerks and CIOs with no appetite for a privacy breach carrying their name.

Quebec municipalities face a parallel but distinct obligation under Law 25, that province's privacy statute, which also binds public bodies and requires a formal privacy impact assessment before any project involving personal information crosses provincial or national borders. A municipality in Gatineau or Trois-Rivières evaluating an AI vendor has to document exactly which flows exist under that assessment. Attesting that everything stays in Quebec is not an option, because for almost no software product is that actually true.

Where Canadian Vendors Fit

The vendor landscape serving this need, tools built by companies that are not simply reselling access to a US model, aware of both FIPPA and Law 25, remains thin but growing. Cohere, based in Toronto, has positioned its enterprise offering around Canadian data handling for some time, primarily for larger federal and enterprise contracts rather than municipal budgets. Smaller entrants are targeting the municipal and provincial public-sector tier specifically, where budgets are lower and procurement cycles longer.

Augure is one of these. The company describes itself as a Canadian AI platform built around Law 25 and PIPEDA compliance requirements, with pricing that starts at C$20 a month for its Pro tier and scales to custom enterprise contracts with SSO and dedicated support. That range puts it within reach of smaller municipal IT budgets in a way enterprise contracts with larger US vendors typically are not. Its product set includes a knowledge-base tool for internal document Q&A and a separate legal-review product aimed at contract and NDA triage, though that legal product is marketed primarily to law firms rather than municipal legal departments.

Whether municipal buyers choose Augure, a competitor, or continue running informal pilots under staff-use policies, the underlying test set by the IPC and BC's OIPC is the same. Institutions need to show, in writing, where the data went and who could have compelled access to it. A vendor's marketing claims support that showing. They do not substitute for it. A documented sub-processor list and a completed privacy impact assessment do the actual work.

The Adoption Gap

None of this has produced a wave of resident-facing municipal AI deployment. Most of what is happening remains internal: staff drafting tools, document summarization, FOI request triage, rather than anything residents interact with directly.

Whether that changes depends less on the technology than on whether municipal councils are willing to put a formal AI governance policy to a vote. Most such councils meet a handful of times a month and have limited technical staff, which leaves the question to IT departments making case-by-case calls instead.

Further detail on how Augure structures data residency and inference routing is documented at augureai.ca.