Every year, Immigration, Refugees and Citizenship Canada (IRCC) uses automated tools to sort through hundreds of thousands of temporary resident visa applications \u2014 visitor visas, super visas, study permits \u2014 before a human officer ever opens the file.<\/p>\n
The department says this speeds up routine cases and helps officers focus on complex ones. It also says the system never refuses an application on its own, and that officers remain in control of every final decision.<\/p>\n
Applicants who are refused, delayed for months, or asked for additional documents say they often have no way of knowing whether any of that automated sorting happened to their case at all.<\/p>\n
Researchers at the University of Toronto authored a study<\/a> titled \u201cIs This Not Enough?: Asymmetries in Institutional Accountability and Collective Sensemaking in the Case of Canada\u2019s Algorithmic Visa Triage System\u201d in which, compared IRCC\u2019s own Algorithmic Impact Assessment for its temporary resident visa triage system against thousands of posts from Canadian immigration subreddits, where applicants trade advice and speculate about why their files were handled the way they were.<\/p>\n The paper\u2019s authors found that while IRCC\u2019s documentation emphasizes transparency, procedural safeguards and human oversight, applicants are left to find answers about what happened to them through guesswork and peer comparison \u2014 piecing together bank balance thresholds, property ownership requirements, and processing timelines from other people\u2019s outcomes rather than their own.<\/p>\n The researchers describe this as an \u201cepistemic asymmetry\u201d: institutional interpretability, on paper, does not translate into anything applicants can actually use.\u00a0<\/p>\n A second pattern, which they call jurisdictional asymmetry, tracks something applicants raise constantly online \u2014 whether nationality, the specific visa office or travel history quietly shapes how a file is treated, even though IRCC frames the system as uniform.\u00a0<\/p>\n A third pattern captures how processing delays disrupt weddings, pregnancies, job offers and family reunification in ways the department\u2019s own risk assessment rates as, at most, moderate and reversible.<\/p>\n IRCC said officers are trained to assess every application \u201cfairly and consistently\u2026 regardless of country of origin,\u201d and that its automated tools are \u201cprimarily rules-based\u201d, operate under human oversight and do not make or alter decisions.\u00a0<\/p>\n The department pointed to its Algorithmic Impact Assessments, AI Strategy and public AI Register as evidence of its transparency, and said refused applicants receive written reasons and can request reconsideration.<\/p>\n IRCC also said officers are not permitted to use generative AI tools when deciding an application \u2014 a claim the Toronto Star tested<\/a> in March, reporting that a permanent residence applicant was refused after a generative-AI-assisted review fabricated details of her job duties.<\/p>\n Immigration lawyers who work with refused and delayed clients say that response, while accurate on its face, doesn\u2019t answer the question their clients actually ask.<\/p>\n \u201cThere is no formal mechanism for an applicant to challenge an internal triage classification itself,\u201d said Yameena Ansari, managing lawyer at Ansari Law. Applicants, she said, are rarely told how their file moved through the system, and there is no simple notation in the records they can access indicating whether automated triage played a role.<\/p>\n Zeynab Ziaie Moayyed, a certified specialist in citizenship and immigration law, put it more bluntly: in years of reviewing refusal files, she says she has seen risk-indicator language surface in officers\u2019 notes only a handful of times among thousands of cases.\u00a0<\/p>\n She points to internal IRCC figures suggesting officers spend under two minutes reviewing each temporary resident application, and argues that once a file is flagged as high-risk, officers become more likely to fall victim to automation bias and search for reasons to refuse it.\u00a0<\/p>\n For most clients who are refused despite a complete application, judicial review in Federal Court is the only route left \u2014 a costly, adversarial process that only succeeds because, she says, \u201ca human being has to look at the application and the refusal and be able to justify the refusal,\u201d something she argues rarely happens meaningfully further upstream.<\/p>\n Cassandra Fultz of DF Immigration frames the problem as one the Supreme Court has already addressed, just not in this context. In Vavilov, the Court held<\/a> that a reasonable administrative decision must be transparent, intelligible and responsive to the evidence \u2014 not \u201cboilerplate.\u201d\u00a0<\/p>\n Fultz says she routinely sees refusals that state only that an applicant\u2019s travel purpose \u201cis not consistent with a temporary stay,\u201d with no engagement with the evidence submitted. Whether that language originates with a human or is shaped<\/a> by automation, she said, is almost beside the point if it fails the Vavilov standard either way.<\/p>\n This isn\u2019t the first time Ottawa has faced this exact question. In 2022, a parliamentary committee investigated why study permit approval rates for French-speaking African applicants had fallen<\/a> as low as 27 percent, concluding that racial bias was a factor. By 2023, IRCC\u2019s own figures<\/a> showed that rate had climbed to 34 percent \u2014 still trailing the 38 percent approval rate for English-speaking African applicants over the same stretch.<\/p>\n