Fact-based journalism that sparks the Canadian conversation

Articles Business Environment Health Politics Arts & Culture Society

Special Series Hope You’re Well For the Love of the Game Living Rooms In Other Worlds: A Space Exploration Terra Cognita More special series >

For the Love of the Game

In Other Worlds: A Space Exploration

More special series >

Events The Walrus Talks The Walrus Video Room The Walrus Leadership Roundtables The Walrus Leadership Forums Article Club

The Walrus Video Room

The Walrus Leadership Roundtables

The Walrus Leadership Forums

Subscribe Renew your subscription Change your address Magazine Issues Newsletters Podcasts

Renew your subscription

The Walrus Lab Hire The Walrus Lab Amazon First Novel Award

Amazon First Novel Award

As Ottawa Gambles on AI, Canadians Live with the Mistakes

Federal workers are being pushed to use tools that can turn made-up facts into official decisions

Last year, Kémy Adé, a post-doctoral research fellow and guest teacher at McMaster University, applied for permanent residence to Canada. Her application made her, in the words of her immigration lawyer, “a prime candidate.” Adé is a health scientist specializing in immunology of aging—precisely the highly skilled immigrant Canada competes to attract.

Nearly half of Canadian public servants admit to frequently using AI in their work

Federal departments use AI without common binding rules or safeguards

Low regulation on AI usage can lead to judgment errors that shape life-changing decisions

But she was rejected. In its refusal letter, Immigration, Refugees and Citizenship Canada pointed to job duties that did not match the experience Adé had claimed, such as wiring control circuits and assembling panels used to operate machinery and robots. None of those duties, however, were hers. Someone or something—or both—got it all wrong.

As the Toronto Star reported, the letter disclosed that generative artificial intelligence had been used in the processing. Like other federal departments, IRCC has adopted AI to vet applications faster, filtering out those that raise flags from those that require less stringent review. IRCC says that a human always gets the final say, but “I cannot comprehend how any human being could make this decision,” Adé’s lawyer told the Star. “Somehow, it hallucinated my client’s job description.”

Adé’s experience is a glimpse of a much larger, scarier—and murkier—experiment under way across Ottawa. AI’s footprint has grown exponentially since ChatGPT’s introduction in 2022, with more than 80 percent of companies globally using the technology. That ubiquity puts enormous pressure on leaders to keep pace. Mark Carney’s government isn’t shirking. As it splashes out on public-facing AI investments and policy, framing it as a necessity for future economic success, federal departments and agencies are availing themselves of the technology for everything from translation to detecting national security threats.

But while private industry has a responsibility to the public at a consumer level, the federal government’s role is different. A corporation can adopt AI to maximize profits first and protect people as a secondary consideration. The federal government can’t, or at least shouldn’t. It is responsible to citizens, and those citizens might rightfully expect the standards to be a lot higher. In theory, they are. But in practice? Well.

For all its vaunted miracles—it can automate routine work, accelerate research, code—AI remains one of the most fallible tools we have. Ontario’s auditor general reported in 2026 that approved AI scribes fabricated patient information in 45 percent of cases, recorded wrong meds 60 percent of the time, and missed key mental health details 85 percent of the time.

The risk is clear. And yet it is astonishing how little clarity there is in how it is being used by federal employees. Some of that adoption is formally declared and governed. Much is not. Any visibility—which tools, for what purposes, on what data—varies by department or may not exist at all. This opacity is partly structural. Canada’s federal government is balkanized, with each department effectively operating as a separate kingdom. Each deputy minister may have their own chief information officer and their own appetite for AI. There is no central strategy to constrain these things and channel them down the same path. One person with knowledge of the overall AI rollout told me that, from one department to the next, “it’s really hard to say what’s happening inside the stack.”

In cases where the government admits AI was involved, as it did with Adé, it has offered assurances that a human reviewed the work and made the final call. But those assurances rarely hold up. Then, there’s the black box nature of the technology itself. Commercial confidentiality rules protect the tools Ottawa purchases from outside scrutiny, making it difficult to know how they reach their conclusions. The result is a troubling gap between what we deserve to know about a government that seems to lean so heavily on AI and what it can—or will—tell us about how those systems work.

The promised revolution has so far produced remarkably little worth celebrating. Maybe you’ve heard about Ontario being sued for allegedly using an AI-powered prisoner classification system that unfairly assigned Black prisoners to higher-security facilities. No? How about Project UNIR, Quebec’s massive effort to hand social assistance services over to automated systems, which led to one man taking his own life after receiving incorrect information—suicidal ideation is now widespread among users. Or the algorithm that was supposed to modernize Canada’s food-safety system; a deadly listeria outbreak slipped through anyway.

It would be nice to say the government has its arms fully wrapped around AI’s march through the federal bureaucracy or even that it understands the risks as clearly as it recognizes the economic promise. But you can’t.

It’s not as if AI took Canada’s federal government by surprise. Far from it. In fact, Ottawa was ahead of the game in some respects, developing strategies around AI tools for at least a decade before the generative boom. In 2016, the government gathered research from hundreds of computer science experts and officials to create a white paper on responsible AI use. That led, in 2019, to the Directive on Automated Decision-Making, or DADM, which imposed requirements for federal use of automated systems developed or procured after 2020.

The DADM remains the only binding federal AI policy, but its reach is limited. The directive applies whenever federal departments use technology to make—or help humans make—a range of decisions, such as determining eligibility for benefits or permits, who can enter Canada, public appointments, or authorizing pharmaceutical products. Before such a system goes live, officials........

© The Walrus