Yes, in every accepted Canadian framework for automated systems — what changes is how much, based on what the agent is allowed to do.
Short answer
Yes. No Canadian framework for automated decision-making — government or voluntary industry code — endorses leaving an agent to run completely unwatched. What counts as adequate supervision scales with how much impact the agent’s actions can have.
The most detailed Canadian description of accountable automated-system oversight is the Treasury Board’s Directive on Automated Decision-Making. It requires, among other things, “Providing a meaningful explanation to clients of how and why the decision was made” after any decision an automated system makes or assists, alongside advance notice that a decision will involve one.
That directive binds federal government departments, not private Canadian businesses — it is a real scope trap, and a business is not in breach of it for running an unsupervised agent. It remains useful, though, as the clearest Canadian articulation of what accountable oversight looks like in practice: notice before, an explanation after, and a documented review of impact before deployment. That review has to happen before the system ever runs, not after something has already gone wrong — the directive requires “Completing, approving and publishing the final results of an algorithmic impact assessment… prior to the production of any automated decision system”, which is the government’s own version of deciding how much supervision a system needs before, not after, it is switched on.
Canada’s Voluntary Code of Conduct on advanced generative AI — signed by 46 organizations including several Canadian banks and technology firms — commits its signatories to a named outcome: “human oversight and monitoring — system use is monitored after deployment, and updates are implemented as needed to address any risks that materialize”. It is voluntary and binds only its signatories, but it is the closest thing Canada has to an industry norm on ongoing oversight rather than a one-time check.
In practice, match the supervision to the stakes: a scheduling agent might only need a periodic spot-check, while an agent approving refunds or contacting customers directly needs review before, or immediately after, each action. The operating rule worth keeping in mind either way — an AI system drafts, extracts, schedules and summarizes; a person decides and signs. See what human-in-the-loop really means and how to decide what a human must approve.
See how ongoing oversight gets built into a live AI system.
