Canada AIDA AI Controls Mapping: Canadian AI Obligations to Enforcement Points

Canada's AI rulebook is a patchwork held together by a privacy statute written in 2000. The Artificial Intelligence and Data Act would have replaced that arrangement with risk tiers and record-keeping duties, and it died with Bill C-27 when Parliament prorogued on 6 January 2025. What binds a Canadian organization today is PIPEDA, Quebec's Law 25, OSFI expectations for federally regulated financial institutions, and the Treasury Board Directive on Automated Decision-Making for federal departments.

Each of those obligations resolves to the same operational moment, and Canadian governance programmes rarely make the connection explicit. I want to set out the mapping obligation by obligation, from the requirement to the control that enforces it where personal information actually leaves the network.

The control point: one HTTPS request

Every mapping below lands on an HTTPS POST carrying a context window to a model endpoint. That request is where the caller's identity is known, where the classification of the prompt can be evaluated, where the destination region is chosen, and where a decision can be recorded or refused. A control that lives in a governance framework describes the obligation and enforces nothing.

PIPEDA Principle 4.1: accountability

Schedule 1 makes the organization responsible for personal information under its control and requires it to demonstrate that responsibility. The control is a per-request record capturing identity, prompt classification, destination, and decision for every AI call. Enforcement produces the record as a byproduct, which converts a stated commitment into something a reviewer can query.

PIPEDA Principles 4.3 and 4.7: consent and safeguards

Principle 4.3 requires knowledge and consent for collection, use, and disclosure. Principle 4.7 requires safeguards appropriate to the sensitivity of the information. Sending a customer record into a third-party model is a use and often a disclosure. The control is prompt-level classification tied to per-role policy, so a caller whose role has no basis for sending health or financial data into a model gets blocked before transmission. The evidence is the per-request classification and the policy decision attached to it.

Quebec Law 25, section 12.1: automated decisions

In force since 22 September 2023, section 12.1 requires informing an individual of a decision based exclusively on automated processing, supplying the personal information used and the principal factors behind the decision, and offering review by a person. The control is a per-decision record binding input to output and the applied policy, plus a route flag marking decisions that must support human review. This parallels GDPR Article 22 handling, and the evidence is the decision lineage a Quebec resident's request lands on.

Quebec Law 25, section 3.3 and cross-border communication

Law 25 requires a privacy impact assessment before communicating personal information outside Quebec, and PIPEDA's accountability principle requires comparable protection when a processor sits abroad. The control is destination-aware routing: prompts carrying regulated classifications reach only endpoints in permitted regions or under an approved arrangement, or they get blocked. AI data residency controls implement this, and the evidence is a per-request destination-and-classification log.

OSFI model risk and the Treasury Board Directive

OSFI Guideline E-23 sets model risk management expectations for federally regulated financial institutions, and the Treasury Board Directive on Automated Decision-Making requires an Algorithmic Impact Assessment plus ongoing monitoring for federal systems. Both assess a system before it runs. The control that covers the operational half is continuous access governance on the model channel: per-role, per-route policy replacing shared static keys, failing closed on ambiguity. The evidence is a running access record rather than a point-in-time score.

The mapping at a glance

The breach-reporting row

One row only activates after something goes wrong. PIPEDA requires reporting a breach of security safeguards to the Privacy Commissioner and notifying affected individuals where the breach creates a real risk of significant harm. When the AI channel is the exposure path, you cannot assess risk on traffic you cannot reconstruct. The independent per-request record turns that assessment into a timeline query, and it is the same artifact behind the Canada AIDA audit-evidence view. My frank opinion: the real-risk-of-significant-harm test is unanswerable without prompt-level records, and organizations discover that during the incident rather than before it.

DeepInspect

Each row above resolves to an enforcement point, and DeepInspect is that point. It sits inline between your users or agents and the LLM APIs they call, evaluates identity, classification, and destination on every request, enforces per-role and per-route policy, and writes a signed record of the decision before the response returns. The Canadian patchwork stops being a set of principles spread across four instruments and becomes one running control with evidence attached.

For an organization that paused its AI governance work when AIDA collapsed, that means PIPEDA accountability and section 12.1 both have a per-decision record standing behind them. Book a technical deep dive at deepinspect.ai.