Canada’s federal, provincial and territorial privacy commissioners are the only Canadian regulators to use the term in published guidance, and they use it to draw a line between two roles. Their generative-AI principles define “Developers and Providers” as “Individuals or organizations that develop (including training) foundation models or generative AI systems, or that put such services onto the market.”
The same guidance spells out what that means in practice: developers and providers are “those organizations that determine how a generative AI system operates, how it is initially trained and tested, and how it can be used”, as against “Organizations using Generative AI”, defined as “Organizations (or individuals acting on behalf of an organization) using a generative AI system as part of their activities.” A foundation model is what the first group builds: a large model trained once on broad data, before anyone narrows it to a specific job. Almost every Canadian business that “uses AI” is using someone else’s foundation model, not building one — which is exactly why the regulators split the two roles the way they did, and why the same organization can, in their words, “shift between or play multiple roles at once.”
The relationship to a large language model is one of scope, not substance. A large language model is a foundation model built for text; the same underlying idea — train broad, then adapt — also produces image, audio and code models. The Cyber Centre’s own guidance describes LLMs as “a subset of generative AI” for exactly this reason.
A general-purpose foundation model is trained by one company on a broad mix of public data. A bank licenses it and, through fine-tuning on its own past customer-service transcripts, narrows it into a tool that only ever answers questions about the bank’s own products. Under the OPC’s role split above, the company that built and trained the original model is squarely a developer; the bank starts out as a user of someone else’s system and moves toward the developer side the further it goes in retraining it — which is precisely the shifting-roles problem the guidance is written to anticipate.
See also large language model, generative AI and fine-tuning.
This is one term in a plain-English glossary on how AI actually works and where it fits in a Canadian business.
