Canada Updates Its AI Register After Public Consultation

Published · 10 min read

Canada’s AI Register MVP is being replaced. Here is what consultation feedback signals for federal AI governance and procurement documentation.

Canada Updates Its AI Register After Public Consultation

What The MVP Replacement Suggests For Federal AI Governance And Procurement Readiness

Federal AI work is moving from internal experimentation toward public, structured disclosure. That shift matters for any organization that builds, integrates, or operates AI systems for Government of Canada departments, because the supporting documentation increasingly needs to stand up to publication, not only internal review.

Canada.ca describes the Government of Canada (GC) Artificial Intelligence (AI) Register as a tool that “collects and publishes information about the GC’s AI systems.” The first release was an MVP, designed to gather feedback on what the public and partners need next. The feedback now points to what the replacement register is likely to demand in practice: consistent fields, clearer risk framing, and linkages to existing governance instruments that already carry operational requirements.

Below is what the AI Register work signals, grounded in what the government has published and what respondents asked to see in the next version.

What The AI Register Is, And What The Government Says Comes Next

In “Consultations on the Government of Canada AI Register: What We Heard,” Canada.ca says the GC Artificial Intelligence (AI) Register “collects and publishes information about the GC’s AI systems.” The same document says the register fulfills a commitment made in the “AI Strategy for the Federal Public Service 2025–2027.”

The “What We Heard” document describes the first release as a “minimum viable product (MVP), or an early version with only basic features and content,” intended to engage partners and the public on the content and functions they would like in future versions. It also states the MVP was assembled from existing sources, including algorithmic impact assessments, access to information requests, responses to parliamentary questions, personal information banks, and the GC Service Inventory.

On “Progress on AI in government,” Canada.ca lists a milestone titled “Publication of the What We Heard Report on the Consultations on the Government of Canada AI Register (Minimum Viable Product) (November 18, 2026).” That page shows “Page details 2026-11-19,” and the “What We Heard” page shows “Page details 2026-11-18.”

Both pages include the same forward-looking commitment: “The feedback collected will be used to develop an improved version of the Register that will replace the MVP and will be updated regularly.”

What The Consultation Feedback Suggests The Replacement Register Will Expect

The “What We Heard” report is not a new rule by itself. It is a requirements signal: a set of recurring concerns and suggested improvements that Treasury Board of Canada Secretariat (TBS) says will inform the replacement register.

From A CSV To A Public-Facing Transparency Product

Respondents said the MVP was “difficult for non-expert users to understand,” and that the “CSV format did not support practical filtering or searching.” They also raised “accessibility and bilingual usability” issues.

Suggested improvements included “Clarify audiences,” “Create an ‘About the Register’ landing page,” “Introduce standardized brief system summaries (‘system cards’),” and “Improve accessibility compliance and bilingual presentation consistency.” Respondents also asked to expand search and filtering so users can filter by:

  • “department or organization”
  • “use case, purpose or type of use (for example, triage, routing, recommendations) or capability”
  • “risk or algorithmic impact assessment impact level or decision type”
  • “affected stakeholder types”
  • “public‑facing versus internal use”
  • “data type (for example, personal, health, financial)”
  • “vendor”
  • “life-cycle stage”

Practical implication for vendors and delivery teams: if the register replacement emphasizes system cards, plain-language summaries, and structured filters (including a “vendor” field), then procurement and implementation work may increasingly need to produce publication-ready descriptions and consistent metadata that can be reused across entries.

Standardization As A Working Constraint, Not An Extra

Respondents flagged “wide variation in the completeness, depth and quality of Register entries,” plus “inconsistent terminology.” They also noted “Status labels (for example, ‘pilot,’ ‘in development,’ ‘retired’) were inconsistently applied.”

Suggested improvements included: “Create a public-facing data dictionary or codebook,” “Use drop-down lists where possible to standardize responses,” “Establish controlled vocabularies for core fields such as system status, AI capability type and data categories,” “Introduce a visible completeness indicator for each entry,” “Define a minimum disclosure standard that all entries must meet,” and “Require last updated dates and a scheduled review cadence for each system.”

A second-order implication for procurement: if “minimum disclosure,” controlled vocabularies, and “scheduled review cadence” become normal expectations for register entries, documentation starts to behave like an operational artifact that must stay current. That can shift effort from one-time implementation deliverables toward repeatable update workflows, with clearer ownership over who maintains the record over the system life cycle.

Risk And Governance Linkage To The Directive On Automated Decision-Making

Several suggestions focus on linking the register to existing governance artifacts. Respondents asked to “Introduce risk-tiered classifications (low, medium, high impact), aligned with the Directive on Automated Decision-Making.” They also suggested to “Require explicit linkage to the system’s algorithmic impact assessment level and documentation or a clear justification if exempt,” and to “Add links to other governance documents such as privacy impact assessments, personal information banks, and the GC Service Inventory.” Another suggestion: “Visually differentiate high-impact systems within the Register interface.”

Canada.ca’s “Algorithmic Impact Assessment” page defines four “Impact level definitions” with labels and score percentage ranges:

  • “Level I, Little to no impact, 0% to 25%”
  • “Level II, Moderate impact, 26% to 50%”
  • “Level III, High impact, 51% to 75%”
  • “Level IV, Very high impact, 76% to 100%”

The same page states: “Impact levels determine the mitigations required under the Directive on Automated Decision-Making.” It also states: “Under the directive (subsection 6.1.3), departments are required to review, approve and update their published AIAs on a scheduled basis.”

The “Guide to Peer Review of Automated Decision Systems” adds a publish-before-production requirement for certain projects. It states: “Projects assigned an impact level of 2 or higher are subject to the peer review requirement (subsection 6.3.7) that mandates publication of a complete review or plain language summary prior to the system’s production.” The guide also states: “Projects assigned an impact level of 2, 3 or 4 must undergo peer review.”

What this signals for suppliers: as the register becomes more explicitly connected to the Directive on Automated Decision-Making artifacts, teams may need to plan for linkable, publishable outputs earlier, including plain-language summaries that can stand on their own outside a technical procurement context.

Clearer Descriptions Of How AI Relates To Decisions

Respondents said the register did not clearly explain whether systems are “advisory, partially automated or fully automated,” and they flagged vague terms like “AI-assisted.”

Suggested improvements include: “Introduce a mandatory field that clearly identifies whether an AI system supports decision-making,” “Introduce a mandatory, categorical field clearly classifying the AI system’s decision role (for example, advisory, human in the loop, fully automated),” and “Add a short plain-language description explaining where the AI system sits in the decision workflow and how its outputs are used.”

Procurement relevance: categorical disclosure forces clarity. If an implementation relies on informal human review practices, those practices become part of what must be described and defended, because the register is oriented toward how decisions happen, not only which tools exist.

Data, Residency, Vendor Role, And Dependencies As Publishable Fields

The “What We Heard” report says respondents wanted stronger data-use transparency. It notes gaps such as not clearly indicating whether “personal, sensitive or protected data were used,” and it says “Data residency, vendor involvement and third-party system dependencies were not always transparent.”

Suggested improvements include: “Standardize disclosures on data categories (for example, personal, health, financial, biometric),” “Clearly indicate data residency (in Canada or outside Canada),” and “Identify the role of vendors (tool provider, operator, co-developer).” The report also includes: “Where disclosure is limited by contract or security restrictions, include a transparency limitation statement.”

One non-obvious implication: once vendor role and dependencies are fields meant for filtering, they stop being procurement-only details and become public descriptors. That can change how teams write statements of work and how they document boundaries between tool provider, operator, and co-developer responsibilities.

Recourse, Accountability, And The Operational Cost Of Being Findable

Respondents called for mechanisms enabling individuals to “understand whether AI was involved,” “seek recourse,” and “enquire further.” Suggested improvements include: “Require each entry to include a named point of contact (or functional role) and a description of how individuals affected by AI-assisted decisions can seek clarification or redress,” plus “Link each high-impact system entry directly to the relevant appeal or redress mechanism.” Respondents also suggested: “Add a feedback and incident reporting feature to allow users to flag AI harms or system failures,” and “Include a mechanism for affected individuals to flag that they believe an unlisted AI system was used in a decision affecting them.”

Operationally, this pushes beyond documentation. Systems that are publicly described and searchable may drive real inbound questions and reports. Delivery teams may need an intake pathway, ownership for responses, and a consistent narrative that matches what is published in the register entry.

How Agentic AI Guidance Raises The Bar For Procurement Readiness

On “Progress on AI in government,” Canada.ca lists “Publication of the Guide on the Use of Agentic Artificial Intelligence (May 22, 2026).”

The “Guide on the Use of Agentic Artificial Intelligence” says it provides “practical direction for assessing, managing, and governing these AI systems.” It also states: “If you are considering using agentic AI systems, always consult on whether the Directive on Automated Decision-Making applies.”

Procurement is explicitly in scope. The guide says: “Aside from research and personal use that do not affect clients, when you use, pilot, procure, or deploy agentic AI, you should always consult with experts early in the project (such as privacy, security, legal, IM, program owners)…”

It also includes a “You should consider agentic AI only when:” list, including: “intended outcomes are clearly defined,” “decision boundaries are explicit,” “accountability is clearly defined and designated,” and “risks can be tested, monitored and managed across the system life cycle.”

For vendors, the signal is practical: proposals and delivery plans may increasingly be evaluated on whether they include early cross-functional engagement, explicit decision boundaries, and life-cycle monitoring plans that match how the GC guidance frames agentic systems.

What Leaders Who Sell Into Government Should Do Next

This is a documentation and operating-model issue as much as a technical one. If your work touches federal departments, a useful internal test is whether your system story can survive being turned into a register entry with structured fields and plain-language summaries.

Actions that follow directly from the published direction and consultation themes:

  • Prepare a plain-language “system card” draft for each AI-enabled system you support, including the system’s purpose, where it sits in the decision workflow, and how outputs are used.
  • Map your documentation to the labels and artifacts the GC already uses, including Algorithmic Impact Assessment impact levels (Level I to Level IV) and peer review publication expectations for projects assigned an impact level of 2 or higher.
  • Decide who owns ongoing updates, aligned to requests for “last updated dates” and a “scheduled review cadence” for each system.
  • Write a crisp, defensible statement of vendor role (tool provider, operator, co-developer) and key third-party dependencies, because respondents asked for those fields to be transparent and searchable.
  • If you support agentic AI use cases, structure the project so decision boundaries and accountability are explicit, and plan for early engagement with privacy, security, legal, IM, and program owners, consistent with the agentic AI guide’s direction.

For teams without a dedicated policy function, this is where a research and documentation sprint can pay off. IN-STORE’s Research service can help translate these federal signals into a repeatable set of system narratives, disclosure-ready field values, and publishing workflows that match how departments are being asked to present AI systems publicly.

The AI Register work points to a simple reality: federal AI delivery increasingly includes public-facing disclosure, standardization, and ongoing updates. Teams that treat documentation as an operational product, not a final report, are better positioned for what the replacement register is being shaped to do.

If you want help turning your AI system information into structured, plain-language materials your team can maintain over time, explore IN-STORE Research and documentation support.

Contact IN-STORE to discuss research, documentation and publishing readiness for federal AI projects.

Sources:

More articles from IN-STORE · This article as Markdown · RSS