Canada's Algorithmic Impact Assessment Registry — Mandatory Pre-Deployment Disclosure for Federal AI

Canada's Directive on Automated Decision-Making, issued 4 March 2019 and in force from 1 April 2019, requires federal departments to assess, mitigate and disclose the impacts of automated decision systems used in administrative decisions about the public, centred on the Algorithmic Impact Assessment (AIA) - a questionnaire scoring risk on a four-level scale.

Ensure departments assess and publicly disclose the risks of automated decision systems before they go into production, for systems scored Level II and above.

Departments completed AIAs covering systems including automated triage tools for immigration applications, the ArriveCAN proof-of-vaccination recognition system, and disability-benefit decision support. A 2023 update added requirements to assess impacts by gender and age. Since November 2025 the public register has held individual records for every federal automated decision system.

As of August 2024, Canadian federal agencies had published 22 AIAs. A February 2025 review by the World Privacy Forum found the framework had been revised multiple times but flagged expertise shortages inside departments, difficulty designing risk-scoring questions, and publication requirements 'often met only in a partial manner.'

Six years in, the directive stands as one of the most mature statutory frameworks for disclosing government AI use before deployment, while its own external reviewers document that coverage and rigour remain uneven across departments; the review's own conclusion was candid that whether the AIA approach 'works as intended' remains an open question.