Public Sector · Compliance

Canada's Automated Decision-Making Directive: What Public Sector Vendors Must Do Before June 2026

The Treasury Board's Directive on Automated Decision-Making has a hard compliance deadline approaching in June 2026. Every federal automated decision system — including those procured before the latest amendments — must be fully compliant. Here's what public sector teams and their vendor partners need to know and do right now.

  • 6 min read
  • Moneli Insights
On this page

The deadline: June 24, 2026

The Treasury Board of Canada Secretariat updated the Directive on Automated Decision-Making to require that all existing automated decision systems — not just new deployments — achieve full compliance by June 24, 2026. This includes systems procured or deployed before the directive's original publication in 2019. If your organization uses Power Automate workflows, AI-driven case routing, chatbots that triage public inquiries, or any software that influences decisions affecting individuals' rights or interests, this deadline applies to you.

What counts as an "automated decision system"?

The directive uses a broad definition. An automated decision system is any technology that either makes or assists in making an administrative decision about a person. This includes:

  • Fully automated systems — decisions made without human intervention (e.g., eligibility screening, automated denials)
  • Decision-support systems — tools that recommend, score, rank, or flag cases for human review (e.g., risk-scoring models, priority queues)
  • AI-powered intake and triage — chatbots or workflows that route inquiries and determine service levels

If your system touches a decision that affects someone's access to a government service, benefit, or process, it likely falls under the directive.

Step 1: Complete an Algorithmic Impact Assessment

The Algorithmic Impact Assessment (AIA) is the cornerstone of the directive. It's a structured questionnaire that evaluates your system's potential impact on individuals and communities. The AIA classifies your system into one of four impact levels:

  1. Level I — Little to no impactInformational tools, analytics dashboards
  2. Level II — Moderate impactQueue prioritization, case routing
  3. Level III — High impactBenefit eligibility, risk assessments
  4. Level IV — Very high impactHealth, safety, liberty, or legal decisions

The impact level determines the requirements you must meet — from basic documentation (Level I) to independent audits, peer review, and public disclosure (Level IV). Many organizations underestimate their impact level. A chatbot that denies service or a workflow that deprioritizes certain cases can push your system into Level III or IV territory.

Step 2: Map your human-in-the-loop requirements

The directive mandates human oversight proportional to impact. For Level II and above, a qualified human must be able to review and override any automated decision before it takes effect. For Level III and IV, decisions must include a meaningful explanation to the affected individual and a documented recourse process.

In practice, this means auditing every workflow where automation touches a decision point. Ask: Can a human intervene before this decision reaches the individual? Is the override mechanism documented and actually used? Are staff trained to exercise judgment rather than rubber-stamp the system's output?

Step 3: Audit your existing Power Automate and AI workflows

Many federal departments have accumulated Power Automate flows, Power Apps, and AI models that were deployed iteratively — often without a centralized inventory. Before June, you need to:

  • Inventory every automated workflow that touches administrative decisions
  • Classify each against the AIA framework
  • Document data sources, logic, and outputs in plain language
  • Verify data quality and bias controls — the directive requires that training data and decision logic do not produce discriminatory outcomes
  • Confirm notice and transparency — affected individuals must be informed when an automated system is involved in a decision about them

Step 4: Establish ongoing governance

Compliance is not a one-time exercise. The directive requires ongoing monitoring, periodic reassessment, and documentation of system changes. This means:

  • Assigning a designated owner for each automated system
  • Scheduling regular AIA reviews (at minimum when the system changes materially)
  • Maintaining audit logs for decisions and overrides
  • Training staff on their oversight responsibilities
  • Publishing AIA results for Level III and IV systems

What vendors and integrators should do now

If you're a technology vendor supplying automated tools to the federal government, your clients will be asking you for compliance support. Proactively prepare:

  • System documentation packages — plain-language descriptions of what the system does, what data it uses, and how decisions are made
  • Bias and fairness testing results — demonstrate that your system has been evaluated for discriminatory outcomes
  • Human override architecture — show how your system supports human review at each decision point
  • Transparency templates — pre-built notice text that departments can customize for affected individuals
  • Change management procedures — document how updates to the system trigger re-assessment

The cost of non-compliance

While the directive itself does not specify fines, non-compliance carries significant risks. Departmental audits by the Office of the Auditor General can flag non-compliant systems. Procurement processes increasingly require AIA completion as a prerequisite. And public trust — already fragile around government AI use — erodes quickly when automated systems make errors without proper oversight.

The June 2026 deadline is not aspirational. It is a hard compliance date.

Organizations that haven't started should begin immediately — and those already underway should verify their assessments are thorough, not superficial.

A practical 90-day checklist

  1. Week 1-2Inventory all automated decision systems across your department or product suite
  2. Week 3-4Complete or update Algorithmic Impact Assessments for each system
  3. Week 5-6Identify gaps in human oversight, documentation, and transparency
  4. Week 7-8Implement remediation — add override mechanisms, update notices, document logic
  5. Week 9-10Train staff on oversight responsibilities and recourse procedures
  6. Week 11-12Final review, publish AIA results where required, establish ongoing monitoring

Keep reading

More from the blog

Practical guides on AI automation, procurement, privacy, and operations.

All articles