“It is difficult to imagine how such a regulatory strategy will inspire greater trust in most Canadians.”
Blair Attard-Frost
Canada CIFAR AI Chair; Fellow, Alberta Machine Intelligence Institute; Assistant Professor, Department of Political Science, University of Alberta
Most Canadians do not trust AI and are not excited to use AI.
The 2026 edition of the Ipsos AI Monitor published earlier this week shows that 67 percent of Canadians feel nervous, and only 26 percent of Canadians feel excited about it. This follows a 2025 survey by KPMG and the University of Melbourne showing only 34 percent of Canadians are willing to trust AI, while 75 percent of Canadians expect our government to regulate AI.
The government states that “trust is the North Star of this strategy” but offers limited protections in pursuit of that North Star. The strategy’s regulatory framework for AI is largely voluntary. While proposed legislation for data protection and online safety from the Trudeau era will be revived in some form, the Carney government does not commit to put any type of high-risk AI system on a statutory footing. Instead, this strategy proposes that a suite of voluntary certification schemes, standards, and “proactive” engagement with AI companies will be sufficient to protect Canadians against the well-documented harms frequently caused by AI systems.
This will be a loose, inconsistent, and opaque regulatory regime. The extent of protections that Canadians receive will be decided not through a democratic process or independent regulatory body but through market forces and closed-door deals with US tech titans. Enforcement will be subject to the motives of political and business leaders rather than consistent, codified law and public oversight mechanisms.
It is difficult to imagine how such a regulatory strategy will inspire greater trust in most Canadians. Though specifics will emerge as the strategy is rolled out, it does not currently appear that AI companies will be required to comply with any certification, watermarking, or other assurance measures described in the strategy. The government will have no statutory obligation to publicly disclose the actions, audits, evaluations, or decisions made in their safety investigations.
While the Artificial Intelligence and Data Act proposed under the Trudeau government had many flaws, it did specify some clear requirements for AI companies along with mandates for public disclosure. Instead of sending a clear signal of regulatory intent, this strategy leaves concerned citizens and businesses with a non-committal scaffold for maybe-stronger-regulation-later. Provinces and sectors will fill this void with their own policies, standards, and guidelines, further fragmenting Canada’s regulatory landscape and creating barriers for interoperability and adoption.
This strategy is an ambitious attempt to connect the dots between a constellation of policy issues in one stroke. Programs for AI literacy, training, infrastructure, and adoption will succeed only if Canadians are excited to participate in those programs and trust our government institutions to protect us against harmful technologies. In safety-critical sectors such as aviation, food, and nuclear energy, regulations that are clear, targeted, and consistently enforced are vital for securing the consumer trust needed for businesses to be able to innovate and adopt new technologies.
AI is also a safety-critical sector. Minister Evan Solomon would be wise to rethink his regulatory approach for this sector, and to learn from the failure of the Artificial Intelligence and Data Act in doing so. Trustworthy, effective AI regulation must begin with deliberative public engagement and civic dialogue. Canada’s anxiety and mistrust in AI are culturally rooted and must be addressed at the cultural level. Public awareness campaigns to build civic AI literacy, a national town hall series, or a white paper soliciting public feedback on proposed legislation would all be good next steps.
“This strategy leaves people on the receiving end of AI systems unprotected.”
Maroussia Lévesque
SJD Candidate in AI Governance, Harvard Law School; Former Senior Policy Analyst, Global Affairs Canada
Let’s start with the positive. The strategy’s push for talent attraction and multinational alliances is a welcome development. In terms of talent attraction, it’s no secret that some of the brightest AI minds trained in Canada, only to seek greener pastures in the US. Canada is spot on to capitalize on the current Move Away, Go Abroad movement (US citizens and residents relocating in light of the political situation) to retain, attract, and regain talent. The move to spearhead multinational alliances with pooled research, talent, and compute also draws from our long-standing tradition of coalition building, including with the Global Partnership on AI.
It’s the smart move in the current political climate, and a good implementation of Carney’s vision for a new world order requiring deeper and broader bonds across middle powers, though some partnerships with Middle East countries with a poor track record on civil and political rights warrant caution.
Next, mixed: measures to support Indigenous leadership and to push for AI in the health care sector. The strategy acknowledges the importance of Indigenous agency and pledges to support self-determination, something Abundant Intelligences and other groups have called for. But contrary to other aspects of the strategy with specific commitments in the millions and even billions, these measures lack specific funding to guarantee implementation. If the government is serious about reconciliation, honouring its treaty obligations, and changing the tides of broken promises, it needs to put its money where its mouth is with hard commits backed by numbers.
As for health care, the strategy proposes to make it the first sector for adoption at scale. Though AI can certainly improve access and care, technosolutionism alone will not “solve” long-standing access inequities and systemic racism.
Incomplete regulation and unambitious compute plans are most problematic. On regulation, the strategy focuses on privacy and protecting children. As important as they are, these selected rights are far from the comprehensive approach AI calls for. Our legal framework should offer people and companies a clear, predictable framework ensuring all rights are respected, providing the ability to understand and contest AI-driven decisions and establishing clear responsibility rules in the complex AI supply chain.
Canada took the wrong lesson from the failed attempt to pass AI regulation. Instead of defining comprehensive obligations based on meaningful consultations, this strategy leaves people on the receiving end of AI systems unprotected and companies wondering how to behave.
As for the sovereign compute vision, it is similarly underwhelming. In a nutshell, it copies the American model—electricity- and water-hungry large-capacity data centres to train ever larger AI models and process user prompts. Even with cleaner energy and cooler climates, replicating the hyperscale approach is doomed to take us closer to climate-change tipping points, at a time when Canada is literally on fire.
Instead, why not rethink a greener AI stack: smaller, purpose-specific models, powered by hyperlocal data centres. In short, replace the “Bigger Is Better” paradigm with a “Small Is Beautiful” mindset. Co-developing a greener compute infrastructure and model architecture is just the first step in crafting an AI strategy that aligns interventions across all layers of the stack.
Overall, the strategy skews toward innovation and casts aside pressing rights and environmental concerns. For a country whose calling card has long been human rights in the international arena, there is untapped potential to anchor the strategy in that tradition. I truly believe we have all the assets to build prosperous, rights-promoting, greener environmental systems. The strategy being a dynamic process, a future iteration can mobilize our strengths toward a more ambitious vision for the future of AI in Canada and beyond.
“This meticulously crafted document is impressive on the surface but light on specifics.”
Sheldon Fernandez
Former CEO of DarwinAI; AI strategist
I’ll admit that the temptation to summarize our government’s national strategy on artificial intelligence using AI itself was not insignificant: a looming deadline, thousands of words to process, and an impatient seven-year-old hovering in the background—it is what generative platforms were designed for. Yet, irony aside, querying a chatbot for its opinion on our country’s long-awaited strategic plan around AI felt limiting and counterproductive.
And therein lies the tension at the heart of this meticulously crafted document—impressive on the surface but light on specifics.
The strongest aspect of the report is the government’s diagnosis of the nation’s position and the challenges it entails. When it comes to AI, Canada is indeed “structurally advantaged, but lagging in adoption and over-exposed to foreign and economic political powers.” Our resources for AI development are formidable, but key infrastructure sits outside the country. As for the AI stack, we do have “presence at every layer, but meaningful dependencies at several”—insofar as cloud, compute, and semiconductor manufacturing services are mostly non-Canadian. It’s a concern but also an opportunity moving forward.
It is also hard to find fault with the government’s goals around privacy, trust, literacy, and an AI ecosystem that endeavours to “expand Canadians’ personal and working lives rather than diminish them.” Enviable, laudable, and . . . fantastically complex in ways that cannot be captured by thousands of words or simply willed into existence.
It is on this last point that the gaps in the plan are most pronounced. How do you innovate with AI in a manner that is safe? “Trust is not a brake on innovation,” claims the report, only sometimes it is—a reality I encountered when I was the CEO of DarwinAI and we had to accept the opaque nature of AI systems to effectively implement them. The report calls our country’s stagnation around AI adoption a “translation problem,” but the issue is deeper and much more ingrained, as I detailed on the RBC Disruptors podcast a few years ago: of our dozen clients, only one was Canadian, a reflection of an extremely risk-averse climate that stifles adoption.
A “commitment to dynamism” sounds great, but as I’ve been writing for decades, being an entrepreneur in Canada is far from straightforward, as the “modesty” and “civility” celebrated in our country can be hard to harmonize with the cutthroatedness required in the global marketplace.
And, finally, ensuring AI can be accessed by every citizen is admirable but overlooks the problematic effects of AI on adult cognition, as
explored in my recent article.
Our government’s seriousness on this topic is to be commended—but reconciling these competing priorities is where the real work begins.
“The term ‘human rights’ does not appear even once.”
Cynthia Khoo
Principal Lawyer, Tekhnos Law; Senior Fellow, The Citizen Lab
Prime Minister Carney is setting out to reshape Canada’s sociopolitical and economic landscape such that fundamental pillars core to a functioning democracy are reoriented to revolve around the false god of AI, an act that mistakes industry dogma for science and stakes all of our futures on it anyway.
What the government proposes—given what we know about the AI bubble and the credible reports and mounting signs it will disastrously burst sooner rather than later—is the equivalent of knowing in advance that the 2008 subprime mortgage crisis would happen, then pouring $2.3 billion into underwriting such mortgages while deregulating investment brokerages, hedge funds, and banks. Much like nostalgia, this document is not a strategy. It is perhaps a few notches above a coercive vision board, and the vision presented is bleak.
First, promises with respect to legal reforms and regulations to protect people against the myriad well-documented harms of AI are underwhelming. They appear more haphazardly thrown together than carefully thought through as a cohesive and comprehensive legal framework, being primarily a combination of legislation already long in progress and meant to have been enacted years before now (e.g., modernized privacy; online harms) and specific issues tied to specific types or deployments of AI that have garnered, for good reason, particularly high-decibel news cycles and public outcry (e.g., LLM-based chatbots, sexual deepfakes).
The focus on “transparency”—as opposed to accountability—and “empowering” individuals with technological (and hopefully critical) literacy to “make informed choices” still ultimately places the burden on individuals to protect themselves, despite the massive information and power asymmetry between them and the AI industry.
Above all, the term “human rights” does not appear even once. That is flabbergasting, given that AI-based technologies have prominently resulted in or are built upon human rights violations at every turn. It is also indicative of where this government’s priorities lie, as is the fact that corporate accountability and liability (to the extent there is any), protection, and safety make up only one of six pillars, while elements to drive the adoption and development of AI technologies occupy the remaining five.
Second, in prioritizing adoption above all, the federal government assumes the only important question is “How?”, before we, as a country, have even had the opportunity to meaningfully consider “Should we?”—or, worse, in the face of clear public and expert opinion, answer that question resoundingly in the negative. Repeated assertions that everyone in Canada has a “meaningful voice” in public debates concerning AI, that people be able to “actively shape how AI is used in their lives,” and that the federal AI strategy will uphold “democratic values” ring hollow when the government has demonstrated the exact opposite in every step leading up to the development of this strategy.
There is no contemplation whatsoever that developing AI literacy is precisely what leads many to opt out rather than want to “participate.” By framing the entire protection and safety pillar around “trust,” the government condescendingly makes people’s perception the problem, not the detrimental impacts of the AI industrial complex itself. By forcing AI tools into schools, early-career training, public services, and health care, the government makes one thing clear: refusal is not an option. Just like with the AI companies the prime minister and AI minister are so enamoured with, informed consent is irrelevant, and lack of consent is but a minor inconvenience to be overridden.
“We really must ask ourselves: AI for all . . . but at what cost?”
Danica Pawlick-Potts
Research Associate in Communication and Media Studies, York University; Member of the Alexis Nakota Sioux First Nation
There is a lot to unpack within this strategy—but I will focus on three points of concern. The first is the overarching theme that Canadians not only need but want “safe, reliable, and sovereign AI,” but what is this claim based on? They themselves report that 36 percent of Canadians see AI as harmful, while “half regard AI as a threat to humanity.” There is an implicit message that Canadians cannot be trusted to know what they want when it comes to AI. It is concerning that the dissenting perspective on AI is not seen as something to investigate and better understand but rather to cure by fostering trust through AI literacy.
I am an advocate for AI literacy as a critical tool for both individuals and communities navigating the AI landscape. Back in 2021, my colleague Mike Ridley and I wrote a piece on algorithmic literacy and the important role libraries can play, and so I am pleased to see not only literacy but also libraries as an avenue for these activities within the strategy. However, literacy should not be equated with a kind of propaganda. People who understand AI, its use, and its impacts should also be empowered to pause, to move slowly and carefully in a technological landscape fraught with well-documented harm and risk, and to ultimately refuse engagement if they so choose.
The second is the disconnect between the described commitment to “link new data centre development with clean energy expansion, robust environmental standards,” and current practice. The proposed large-scale Wonder Valley AI Data Centre will not only be powered by natural gas, according to current plans, but was also deemed exempt from environmental impact assessment by the Alberta government. Echoing Chief Sheldon Sunshine, who is calling for a full federal review of the Wonder Valley project and its environmental impact, the time is now for the federal government to show that this strategy is more than words on a page and to intervene to support Indigenous communities fighting to ensure these data centres are not built at the expense of our lands and waters.
Which brings me to my third and final point of concern: while Canadian sovereignty is noted as a key driver of the strategy and fuels the alleged need to build infrastructure such as data centres, completely absent is any discussion—or even acknowledgement—that much of this infrastructure is to be built on treaty land. Further, there is no reference to Indigenous rights and sovereignty in this document. There is a small discussion of supporting “Indigenous leadership in AI,” but it is framed as supporting participation, not as respecting and protecting the sovereignty and rights of Indigenous peoples and their lands.
We really must ask ourselves: AI for all . . . but at what cost?
“Lurking between the lines is an assumption that trust can be built through literacy efforts. I do not think that is going to cut it.”
Vass Bednar
Managing Director of the Canadian Shield Institute
The strategy is strikingly optimistic about adoption while leaving the harder governance questions still unresolved. It takes largely on faith that AI adoption will translate into productivity gains and economic prosperity, but that is not a sure-fire bet across all sectors. So far, applications of AI have had uneven results, and it’s not yet clear what Canada’s economic value capture strategy actually is. Maybe it’s through government investments. But how does it make Canadians richer? Are we going to have higher-paying jobs? Are we going to see more tax revenue? Are we going to have better products offered more cheaply?
The biggest disappointment is the continued absence of a modern privacy law. Canada has now spent six years circling this issue: C-11 died on the order paper; C-27 revived the effort through the Artificial Intelligence and Data Act (AIDA), and then it, too, died. A year later, we are still waiting. It is hard to ask Canadians to trust AI systems when the most basic governance architectures remain hypothetical.
Lurking between the lines is an assumption that trust can be built through literacy efforts—educating people harder may not change their minds if the underlying market is unreliable. So I do not think that is going to cut it. Actual trust comes from responsible governance and reliable products. For instance, I totally trust the airplane that took me to Ottawa the day of the policy drop. I’m not an aviation expert, but the plane is regulated, and it took me where it said it would, when projected.
Right now, we are encouraging people to fly, adding harnesses later, and accepting that sometimes the flight just goes somewhere completely random. Until we see the other laws that will bolster this strategy, like privacy updates and online harms legislation, we cannot know whether the overall framework is credible.
It feels like Canada is treating AI governance as the after care for adoption rather than the condition that makes adoption legitimate in the first place. A digital sovereignty approach would flip that sequence: governance should shape the market from the outset by setting rules for data use, IP ownership, procurement, standards, and compute access so that Canadian adoption builds Canadian capacity instead of simply producing downstream harms and upstream profits for foreign platforms.
Finally, a nugget that made me gasp, because the federal government had not committed to any action around it, was the acknowledgement of surveillance pricing. Flagging it marks an official recognition that AI-enabled markets often use excessive data to offer unique prices. That felt like a refreshing reality check.
“The strategy needed to protect the project of democracy in an age of AI boosterism. It did not.”
Fenwick McKelvey
Professor in Information and Communication Technology Policy, Department of Communication Studies, Concordia University
Hype undermines our very ability to interpret the federal AI strategy. How can we make sense of AI—as the next industrial revolution, the singularity, or smoke show—if the people we trust to explain the technology are the same ones selling it? The strategy needed to protect the project of democracy in an age of AI boosterism. It did not.
Thirty-five years ago, media theorist Andrew Wernick argued we were entering a promotional culture—one in which every message increasingly served as an advertisement for someone or something. AI is the latest example. I spent yesterday scrolling LinkedIn to see the reactions to the strategy, a forgotten habit since Twitter ended for me. My feed was a mix of people really excited about the document. But when I look at those posts, they are from individuals who need to promote the strategy because they benefit from it. What Canada’s past AI strategies have built is a marketing culture for the technology.
Most alarmingly, this hype legitimates AI as a deeply anti-democratic political project. The top minds in AI have all turned into heels, flying in top scholars of retrograde dark enlightenment projects, or rewiring public infrastructure into private enterprise. Ordinary people are ultimately proving irrelevant to this elite and their corporate rules. Data centres, for instance, are at the heart of the AI strategy. One of its big contradictions is that all these sovereign compute centres are going to be filled with American hardware running American software—what is actually making money in the AI economy. All Canada will contribute is the oil and gas burnt to run these data centres (with some help from our taxed energy grid).
Absolutely nothing about data centres seems like a good long-term investment. In my hometown of Saint John, they are going to flatten 200 acres of spruce forest to build one of these boxes, and for the life of me, I cannot tell you how it will make money or what jobs it will bring to the local economy.
There are signs of hope. Canadians do not trust AI, maybe because outside the hype bubble, it is not too hard to see who is going to win in this economy. Hype for the AI strategy, then, has a perverse effect of wanting to create AI for all using a vision of AI built for the wealthy few, gaslighting a genuine national skepticism as being naive or reactionary. Far from it, the political problem of AI is just beginning, not only for tech billionaires but for the elected officials who boost it.
If the AI bubble is beginning to show its cracks, it’s not least because the technology is so often discussed through the language of the hard sell. Just last year, I had the privilege of being part of a national civil society–led People’s Consultation for AI. We are just analyzing the results, but reading the comments, I am struck by ways that organizations from across Canada can discuss AI pragmatically. I do not hear these voices in the strategy.
AI comes at a time of profound change in the habits of our democracy. The loss of many of the ways we tell truths seems to go hand in hand with exuberance that AI is going to change the world. Back home, we would sometimes pour gasoline on a pile of driftwood. We used to call that a bonfire. We now call that an AI strategy.
“The strategy misconceives public trust as a means to an end.”
Alissa Centivany
Starling Centre for Just Technologies and Just Societies; with Dorotea Gucciardo, Alison Hearn, Joanna Redden, and Luke Stark
“There is nothing more difficult to carry out, nor more doubtful of success, nor more dangerous to handle, than to initiate a new order of things.” Five centuries later, poring over Canada’s new “AI for All” strategy, Machiavelli’s observations still ring true.
A successful state-sponsored innovation strategy faces two key challenges: incumbents are often hidebound, myopic, and resistant to innovations that might risk their power, wealth, and influence; and humans, by nature, are often incredulous. The new strategy is unlikely to overcome these barriers—in fact, it risks entrenching them further.
AI is a transformative technology, but in its current form, it is also a distinctly political project, working to shore up and extend the interests of big tech companies, oligarchs, and politicians. The centripetal force of incumbency resists the very innovation this strategy seeks to promote. The government’s handling of AI to this point has favoured industry leaders, exemplified by the AI Sprint taskforce. Despite extolling the need for homegrown champions to address national resilience, digital sovereignty, and economic development, the Carney government’s recent procurement contracts with Trump-allied tech firms like Palantir suggest otherwise.
Silicon Valley is already deeply embedded inside the workings of the federal government. A recent scan we undertook showed over 218 AI applications being trialled and used by the Canada Revenue Agency alone. Many of these are licensed from private companies, whose business interests are opaque even to the civil servants who implement them. The strategy’s touted investments in domestic champions and small and medium-sized businesses may eventually temper some of the ill-effects of incumbency and foreign reliance, but it is difficult to see how the government can remedy the harms of consolidated tech power and wealth anytime soon.
Surprisingly, the strategy disregards that AI often does not provide the outcomes promised. The CRA recalled its chatbot Charlie, for example, because it was wrong 67 percent of the time. Alarmingly, claims about protecting “our children and our citizens, our culture and languages, and our democracy” contradict rapidly accumulating research showing that AI can increase discrimination and inequality and be used in ways that infringe upon rights, disrupt democratic processes, wrongly limit access to services, introduce error, intensify surveillance, and exacerbate the climate crisis. Despite extensive evidence of AI and automated decision systems causing widespread harms, the federal government continues to take actions on AI that concentrate power and profit for the few while socializing the risks and harms caused by AI applications.
Given the strategy’s willful blindness to the “All” in “AI for All,” it should come as little surprise that the overwhelming majority of Canadians do not trust AI and do not want AI integrated into their lives. The government’s answer to this trust deficit is to promote siloed responses related to privacy, online harms, and AI literacy. Broad statements about protecting Canadians from AI risks and harms without articulating details about how safety and security will be made actionable and enforceable will raise more questions and suspicions in the Canadian public, not less.
The submission of over 11,000 public comments during recent consultations underscores significant public interest in democratic participation, yet many of the perspectives raised received little attention or were omitted entirely from the strategy. Evan Solomon’s use of AI to review those public submissions further signals how little this government values public input and human oversight, and how much blind faith it puts into AI systems themselves.
Finally, the strategy misconceives public trust as a means to an end: AI adoption. If we do not adopt AI of our own free will, the government will persuade us through so-called AI literacy and training programs in schools, financial incentives, and mandated integration into administrative processes and public services. Here, again, we see shades of Machiavelli’s end-justifying-means calculus. If entreaties and persuasion fail to garner the sought-after result, “it is necessary to order things so that when they no longer believe, they can be made to believe by force.”
When public trust is pursued through aggressive discourses about AI inevitability and imminent threats to sovereignty and prosperity, public incredulity seems a healthy response.
