The story of the “freedom convoy” protests that engulfed Canada in early 2022 is often told as a domestic drama about vaccine mandates, truckers and downtown Ottawa gridlock. But there was a quieter, more strategic player watching from the wings: Russia. As the convoy rolled on, Russian state media gave the protests a level of attention that dwarfed nearly every other international outlet, even though Canadian public health rules had almost no direct bearing on ordinary Russians. It wasn’t the policy dispute that interested Moscow. It was the raw material. The convoy was a perfect gift for an old playbook: take a polarizing event in a Western democracy, amplify it relentlessly, and use it to deepen distrust, widen fractures and make people feel that their institutions are broken, corrupt or rigged. This is what influence operations have always done, but the tools keep changing. For decades, this kind of warfare was waged with leaflets, radio broadcasts and planted newspaper stories. Then came social media, with bots, troll farms and algorithmically boosted outrage. Now, just as we are beginning to understand the damage done by those platforms, the ground is shifting again. The next front in this quiet war will not only be fought in our feeds. It will be fought in the private, seemingly safe conversations we have with artificial intelligence. And unless we think carefully about this now, we may not recognize how deeply our understanding of the world has been shaped until it’s far too late.
This is not a distant or futuristic concern. Researchers at Oxford University and the Reuters Institute have projected that by 2026, roughly one in ten people around the world will already be using AI chatbots to get their news. Among young adults aged eighteen to twenty-four, that figure jumps to about one in six. Those numbers will almost certainly keep climbing as AI gets woven directly into messaging apps, search engines, phones, cars and smart glasses. The trouble is that AI chatbots do not conjure their answers from some neutral repository of universal truth. They learn from the internet, which is vast, messy and increasingly polluted with deliberately deceptive content. Russian influence networks, for instance, have reportedly flooded the web with pro-Russian material through networks like the so-called Pravda outlets, and researchers have documented leading chatbots repeating those talking points as if they were facts. This has become known as “large language model grooming,” and it works like this: if you can fill the digital well from which AI models draw their water, you don’t need to convince people one by one. You just poison the source, and every conversation that follows carries a little of that poison. The target of influence operations in this new environment may no longer be just people at all. It may be the AI models themselves, because a model trained on slanted information will produce slanted answers for millions of users. And unlike social media, where a tweet or a video can be seen, screenshotted and investigated, conversations with an AI assistant are private. By the time a person brings their views into a public debate, the reasoning may have already been quietly shaped by something that left no public record. There is no timeline to scroll through, no comment section to examine, no way to see what exactly persuaded them.
For Canada, this raises a question that has nothing to do with whether the latest AI gadget is entertaining or useful. It is a question of sovereignty and trust. As things stand, nearly all of the major AI models Canadians routinely engage with are built and owned by foreign companies, many of them headquartered in the United States or China. That does not automatically mean they are malicious, but it does mean that Canada is relying on strangers to explain the world to its citizens. The values embedded in these systems, the sources they prioritize, the way they frame controversial issues, and the subtle choices they make about what to include and what to leave out are all shaped by forces outside Canadian control. This is not a reason to panic or to abandon the benefits of AI, but it is a reason to demand more. Canada can require that AI systems use verified and trustworthy sources when it comes to high-stakes areas like voting, banking, health and public safety. In those domains, an answer pulled from a random website or a Kremlin-linked outlet should be treated as unacceptable. Canada can also build alliances with like-minded countries to strengthen its bargaining power in setting global standards for things like identity, provenance and accountability in machine-to-machine interactions. If AI systems are going to talk to each other, verify documents and act on our behalf, we need rules that make it possible to know where information came from and who is responsible when things go wrong. Intelligence agencies have always had to adapt to new technologies, but they will need new technical capacities to understand not just what is happening on social media, but what is happening inside AI systems and in the private exchanges between Canadians and their digital assistants. Public institutions themselves will need to evolve, so that government services, public health advice and trustworthy information remain accessible to people who increasingly turn to AI instead of websites or phone lines.
But this is not only a task for governments, spy agencies and tech companies. Ordinary Canadians have a role to play, and it is more active than simply waiting for some new regulation to arrive. The first step is recognizing that AI systems are never neutral. A chatbot may sound friendly, objective and infinitely patient, but it has been built by someone, trained on something and optimized for particular purposes. It can make mistakes, reflect biases and repeat falsehoods with complete confidence. That means we need to build a habit of questioning what AI tells us, especially when it touches on current events, politics or matters that could affect how we vote, how we treat our neighbours or how we understand our country. Ask the chatbot for its sources. Go to those sources and check them. Then go beyond them and deliberately look for other points of view, especially ones that challenge what the chatbot just told you. This sounds simple, but it requires a certain kind of mental discipline that does not come naturally to most of us. We tend to trust tools that seem helpful and clear. We want answers, not more homework. But the stakes are too high to treat AI like a magic oracle. This is why Canada has much to learn from Sweden, which has built one of the most impressive civilian defence systems in the world. Sweden has a Psychological Defence Agency that trains ordinary citizens to spot malign influence operations and to understand that they are not just bystanders in an information war. The message there is simple and powerful: every person who lives in this country is part of the defence. They are the ones who notice when something feels off, who stop before sharing a dubious story, who refuse to let themselves be turned against their own institutions. Sweden has made resilience a national habit, not just a government program, and it has done so with a calm, practical determination that Canada could easily adapt.
If there is a warning in all of this, it comes from the recent past. Think about how social media transformed public life. At first, it seemed like a wonderful way to connect with friends, share photos and discover new ideas. Very few people imagined that the same platforms would become engines of polarization, radicalization and foreign interference. Democratic governments were slow to understand what was happening, and when they finally did respond, the damage had already been done. Social trust has eroded, public discourse has become harsher, and democratic societies around the world are still struggling to deal with the consequences. The problem was not just that foreign actors used social media to spread lies. It was that the underlying architecture of the platforms rewarded outrage, speed and division. Institutions were caught off guard because they were still thinking in terms of old media, where a handful of gatekeepers controlled what people saw. The internet and social media flattened those gatekeepers, and nobody was ready to fill the gap. We are now at a similar inflection point with artificial intelligence. The technology is arriving faster than our laws, our norms and our institutions can adapt. We are already seeing how AI can generate realistic fake images, videos and even entire articles that are almost impossible to distinguish from the real thing. We are seeing how chatbots can produce confident, fluent nonsense and how they can be manipulated by hidden actors who understand their weaknesses better than the public does. If we wait until the next election crisis, the next foreign interference scandal or the next collapse in public trust, it will be too late. The whole point of grooming an AI model or seeding a chatbot with false information is that it happens quietly, over time, without any dramatic moment that triggers an investigation. By the time someone notices, the narratives have already seeped into millions of private conversations.
This is all the more urgent because the same fault lines that made Canada a target in 2022 are still there, waiting to be exploited. Four years after the freedom convoy, foreign actors are again paying attention to Canadian divisions, this time around talk of Alberta separatism. Intelligence officials and researchers have warned Canadians to be alert to the narratives circulating online, and that is wise advice. But it is only half of the picture. While we are watching the loud, visible content on social media, we are not watching the quiet, invisible formation of opinions happening inside AI conversations. A person who feels alienated from Ottawa, anxious about the economy or frustrated with the pace of change may turn to an AI assistant for answers. They may ask it about Alberta’s place in Confederation, about whether independence is viable, about whether the federal government is listening to the West. If that AI has been trained on databases that include a steady diet of foreign propaganda and highly partisan content, the answers will nudge that person in a certain direction without ever appearing overtly extreme. The chatbot may not tell them outright that separation is the answer. It will simply frame the issue in a way that makes that conclusion seem more reasonable, inevitable and supported by evidence. That is the new nature of influence operations. They don’t need to win an argument in a public square. They just need to shape the assumptions that people bring into the argument before it even begins. Canada still has time to act, but not much. It needs to invest in domestic AI literacy, demand transparency from the companies that build these tools, forge alliances with allies who share its values and build the technical capacity to defend public institutions in an AI-driven world. But above all, Canadians need to understand that AI is not a neutral oracle. It is a reflection of the messy, contested, often poisoned information environment that produced it. The responsibility for keeping it from turning into a weapon against democracy does not rest solely with governments or corporations. It rests with all of us, in the questions we ask, the sources we verify and the conversations we choose to have with each other instead of with machines that may not have our best interests at heart. This is the challenge of our time, and how we respond to it will shape the future of democratic life in Canada for generations to come.