Glossary · Canada
Directive on Automated Decision-Making (Canada)
Also known as: DADM, Algorithmic Impact Assessment, AIA tool
The Directive on Automated Decision-Making is a Treasury Board of Canada policy that governs how federal institutions use automated systems to make or support administrative decisions. It requires an Algorithmic Impact Assessment before a system is used and applies requirements — such as notice, explanation, bias testing and human involvement — graded by impact level.
- AI regulation
- Canada
- Technical documentation
In one sentence
Canada’s Directive on Automated Decision-Making governs federal use of automated decisions, with an Algorithmic Impact Assessment and graded duties.
Example
Before using a system that triages visa applications, a federal department completes the Algorithmic Impact Assessment, reaches impact level III and ensures that officers make the final decisions.
How it applies
- Algorithmic Impact Assessment (AIA): Before production, the institution completes the AIA, an online questionnaire on the system's design, data, decisions and mitigation measures. The result is an impact level from I (little impact) to IV (very high impact), and the results are published on the Open Government Portal.
- Graded requirements: Depending on the level: peer review, notice to affected people before the decision, meaningful explanations after it, testing for unintended bias, monitoring, training of staff, contingency plans and, at the higher levels, human involvement so that the final decision is made by a person.
- Updates: The directive has been in effect since April 2019 and is reviewed regularly; its latest version took effect in June 2025.
- Technical documentation: The AIA and the explanations to affected people are the core documents. Vendors selling to the Canadian government must supply the information needed to complete them — about training data, model logic and testing.
Directive on Automated Decision-Making vs. the EU AI Act
The directive regulates the government as a user of AI, much like the AI Act's duties for public-sector deployers, and the AIA resembles the fundamental rights impact assessment. It does not apply to private companies, and it is a policy instrument rather than a law.