Canadian AI policy explained, with practical guides for everyday use.
About the publicationCANADIAN AI POLICY & PRACTICAL GUIDES
Work through a 20-point answer comparison, rerun two published code experiments, or check a Canadian AI policy against its source documents. Worked examples and measured results are labelled separately.
Canada’s Algorithmic Impact Assessment is a starting point for scrutiny, not proof that a system is safe. Read the scope decision, supporting answers and peer review together; then identify the evidence behind the score and the consequences for affected people.
Policy, public infrastructure and implementation, read as a connected system instead of isolated announcements.
Canadian AI sovereignty involves more than locating a data centre in Canada. Check five things for your workload: location, operational control, access, continuity and exit. The national strategy describes commitments; a supplier still needs to show which controls and capacity you can actually use.
Read the evidence for Canadian AI sovereignty: follow the workload, not the map
A federal AI pilot needs a decision brief that names its task, data, accountable owner and stopping conditions. The 2025–2027 strategy supplies direction, not permission for a particular deployment. Start with a narrow use and evidence that would justify continuing it.
Read the evidence for The first page of a federal AI pilot should be a decision brief
An AI privacy assessment should follow personal information through the whole system: prompts, retrieved documents, model responses, logs and vendor copies. Record why each copy exists, who can access it and when it is removed. A privacy setting on the chat screen does not answer all those questions.
Read the evidence for Where did the prompt go? Trace an AI system’s personal dataRun the invoice and document-selection experiments from the published data and code, or score two answers in your browser. Methods and limits are included.
More ways to test AI
To check an AI-assisted spreadsheet, verify which rows were included before trusting the total. The worked invoice example totals $170, but duplicates, text amounts and changed labels can make a plausible result misleading. Eight published code cases show which errors the sample checks catch.
Our synthetic selector passed its original eight document cases, then failed four of five additional probes involving bad or contradictory source records. Inspect both sets of inputs and outputs to see what version, audience and citation checks do—and what they leave untested.
To compare AI answers, use the same task and sources, then score evidence, completeness, uncertainty, usefulness and permissions. A fluent answer must still fail if it invents authority to act. The 20-point worksheet below includes two fictional answers, source records and an answer key.
Practical guides to writing a clear request, checking the result and keeping private information out of the wrong account.
All practical guides
To turn meeting notes into an action list with AI, request the task, owner, deadline and supporting words from the notes. Require missing details to stay unresolved. This fictional workshop exercise includes a complete answer key so you can spot invented commitments before using the result.
To fix an AI prompt, identify the specific failure before adding more instructions: missing facts, invented details, the wrong format or an unclear audience. Give the model a bounded source card, repair one instruction and test another example. Better wording cannot supply facts the source never contained.
Before uploading a file to AI, remove information the task does not need and check hidden contents as well as visible text. Share a minimal extract where possible. Connecting an email account or folder grants a different scope of access and needs a separate permissions check.
Read the source documents, inspect the experiment files, or tell us which claim needs another look.
About the work and its limits