The idea sounds intuitive enough: if a technology is potentially harmful to children, restrict their access to it. It's the logic that has driven sweeping bans on minors using social media in Australia, and it's the same logic now being applied to AI chatbots in the United States, the U.K., Canada and beyond.

In Canada, the federal government in June introduced Bill C-34, the Safe Social Media Act, which includes an age restriction preventing children under the age of 16 from having accounts on social media services.

But a growing chorus of researchers, policy analysts and civil liberties advocates are pushing back – arguing that blunt bans aren't just ineffective, they may actively harm the children they're meant to protect.

A recent webinar hosted by the Washington, D.C.-based Information Technology and Innovation Foundation (ITIF) brought together four panelists to unpack the current wave of chatbot legislation and explore more nuanced alternatives. Their conversation raises questions that are just as relevant north of the border, where Canada is simultaneously grappling with its own AI policy moment.

At the centre of the U.S. debate is the GUARD Act – Guidelines for User Age Verification and Responsible Dialogue – which would effectively ban minors under 18 from accessing AI companion chatbots and impose strict age verification requirements on all users.

Juan Londoño, policy analyst at the Cato Institute, described it plainly: "It's the most baffling, because it treats these AI tools and AI companions as an inherently dangerous, no-upside technology. It's saying, okay, we just have to ban kids and minors altogether."

What makes the bill particularly sweeping, Londoño noted, is that it offers no pathway for parental override – a provision that even many restrictive proposals include. A 17-year-old whose parent believes AI tools are beneficial for their household would have no recourse under the legislation.

Similarly, there’s no provision in Canada’s Bill C-34 that allows parents or guardians to provide consent or override the minimum-age ban to let their child use a restricted social media platform. The new Digital Safety Commission to be created by the legislation may grant an exemption to a social media service if the platform itself demonstrates that it has implemented sufficient, adequate safeguards to protect children.

In the U.S., the GUARD Act is also being discussed as a potential addition to a broader AI safety legislative package, which raises further concerns. Becca Branum, deputy director at the Center for Democracy and Technology, warned that bundling the GUARD Act with other proposals carries real constitutional risk: "The GUARD Act is uniquely vulnerable to First Amendment challenge."

A question of speech – and who gets to access it

The First Amendment dimension of this debate often gets lost in the conversation about child safety, but it's one the panelists returned to repeatedly.

Josh Withrow, resident fellow at the R Street Institute, framed it directly: "Even kids have a certain measure of First Amendment right to access general-purpose speech platforms. The First Amendment has always been held pretty consistently across time to place a barrier on what the government can do."

He cited former Supreme Court Justice Antonin Scalia's position that "even where the protection of children is the object, the constitutional limits on government action apply."

Branum added another layer to this, arguing that what's actually being regulated when you restrict emotionally resonant chatbot responses isn't a product feature – it's speech. "Some of the best speech, the speech I most care about, is speech that's emotionally resonant. It's like telling kids they could only watch documentaries, not action movies."

She also cautioned against assuming that courts have given legislators a green light here. "There's been exactly one ruling from a court on the First Amendment status of chatbot outputs, and it was more or less a judge saying, ‘I don't have enough information to say that chatbot outputs are protected by the First Amendment.’ I urge caution to anyone who reads that decision as an opening to do whatever they want."

The age verification problem

Even setting aside the speech debate, the practical mechanics of age verification present significant challenges. Withrow pointed out that the math alone is sobering: "Even the best of these technologies are going to have some error rates. On the scale of the Internet, that's tens of thousands, maybe hundreds of thousands of people who, in order to comply, the only alternative would be for them to supply documentary evidence of their age."

Branum raised equity concerns that she worries get overlooked: "Not everyone has access to a government ID, particularly children, who might not have a reason to have it yet. When lawmakers are mandating these types of 'papers, please' interventions, they're not going to be experienced by people equally."

Londoño pointed to an even more layered problem around what "age-appropriate" content even means. "How does a chatbot tackle the question of Santa Claus? In one household, it's perfectly fine to say Santa Claus still exists. But to a seven-year-old in another household, it might be marginal,” he said.

“Or if you're talking about dieting content - for someone recovering from an eating disorder, it might be dangerous. But for a kid managing diabetes, it could be incredibly informative.”

Enforcing the age verification provision is another problem. Peer-reviewed research, including studies published in the British Medical Journal, show that more than 85 percent of children under 16 in Australia continue to access restricted platforms like Instagram, TikTock and Snapchat despite the federal ban.

Half the children used their own accounts – very few were forced to provide official government ID. Also, many lied about their age and fake accounts are a common method of bypassing the ban.

Not a new story

Branum identified something she calls a "social media hangover" driving much of the current legislative push – policymakers who feel they moved too slowly on social media and don't want to repeat that mistake.

Londoño agreed the comparison is instructive, but thinks it's being drawn incorrectly.

"Banning kids from AI chatbots is like banning them from Google search,” he said. “We were taught how to Google effectively. This is going to be the research of the future – your editor, your copywriter – and you're not going to learn how to use it. We're really leaving kids behind."

Justine Gluck, policy analyst at the Future of Privacy Forum, noted that the state-level picture in the U.S. is actually more nuanced than the federal proposals suggest. After tracking roughly 100 bills introduced this year, she observed that outright bans are the exception rather than the rule.

"States aren't generally moving towards a model that would prohibit minors from accessing companions altogether,” she said. “The newer bills are moving beyond just disclosure and into more substantive regulation on design – things like preventing self-harm content, fostering emotional dependence, or encouraging minors to keep secrets from trusted adults."

What Canada brings to the conversation

Canada is navigating its own version of this reckoning, and some of the tensions look familiar.

In Canada, about 50 percent of children are carrying a smartphone by age 11. The World Health Organization’s Health Behaviour in School-aged Children 2022 National Report, Health of Young People in Canada: Focus on Mental Health, found that 39 per cent to 54 percent of Canadian students in grade 9 to 10 report online contact “almost all the time throughout the day.”

Excessive use is linked to a range of negative associations, including depression and anxiety, behavioral problems, reduced physical activity, disordered eating, sleep disruption and reduced family connectedness, according to the report.

A study published by the by the Public Health Agency of Canada similarly found that the likelihood of Canadian youth reporting positive mental health indicators decreased as screentime increased.

The Information Technology and Innovation Foundation and the Macdonald-Laurier Institute have argued jointly that Canada's challenge with AI isn't a shortage of invention – it's a failure to turn research strength into widespread adoption.

Lawrence Zhang, head of policy at ITIF's Ottawa-based Centre for Canadian Innovation and Competitiveness, put it bluntly: "78 percent of firms not using AI say they cannot see how it would benefit their business. That means Canadian businesses aren't holding back because AI is too expensive; they're holding back because they don't know what AI would do for them."

Applying this lens to youth and chatbot policy raises an uncomfortable question: If Canada were to restrict young people's access to AI tools just as AI literacy becomes a foundational workforce skill, would it be protecting children – or handicapping them?

Zhang and co-author Daniel Castro, vice-president at ITIF, have argued that "in an era when AI will enhance everything from crop yields to cancer detection, the central priority should be to accelerate AI adoption to boost Canada's economic prosperity and quality of life for its citizens, not impose roadblocks to innovation with overly precautionary regulation."

This doesn't mean ignoring genuine risks. But it does suggest the frame matters enormously.

One reason chatbots provoke more anxiety than, say, a search engine, may be rooted in how human brains are wired to respond to them. A recent article from researchers at Université de Montréal and Johns Hopkins University, draws on decades of neuroscience to make a critical distinction: intelligence is not the same as consciousness.

Vanessa Hadid, a postdoctoral fellow in psychology at Université de Montréal and the McGill University Health Centre, said: "In a context of psychological support, the risk is not only that AI may respond poorly, but that it may respond well enough for us to forget that there is no one behind the answer."

Karim Jerbi, professor in the Department of Psychology at Université de Montréal and researcher at Mila, agreed: "Anthropomorphism means attributing emotions, intentions or consciousness to something that behaves like a human. With AI, this reflex can become a trap – it feeds the illusion of being understood and can lead to misplaced trust."

The researchers aren't calling for bans. They're calling for informed use and for users of all ages to understand that a system can be remarkably fluent and emotionally attuned without actually experiencing anything. This may be especially important for young people who are still developing their frameworks for understanding social relationships.

"We know from decades of research that people, even when they know in their bones that they're interacting with a machine, they still respond socially and emotionally. It is inherently a different kind of technology,” said Branum from the Center for Democracy and Technology

So what should policymakers actually do?

The ITIF panelists were asked directly: as we look ahead, what approach would you most like to see?

Branum outlined three priorities: "I hope that the federal government leaves space for states to regulate. Privacy is really the name of the game here – a way that we can protect people in constitutionally durable ways. And I hope they take the First Amendment seriously. What I would hate to see is Congress invest a ton of time and effort into a bill that gets slapped down on First Amendment grounds and then we're left with nothing."

Londoño's concern is the patchwork that age-gating creates across jurisdictions. He pointed to how VPN use is now entering the regulatory crosshairs in content regulation, threatening a second layer of privacy loss in the name of the first protection. He also emphasized education, "not only for parents, but for educators. For schools to be prepared for the AI future."

Withrow advocated for a watching period: "Take some of these more extreme proposals completely off the table and watch how these platforms develop. For every horrible anecdote of how these chatbots can misbehave, you already see the companies troubleshooting how to make their platforms more resistant to negative use. There's a huge incentive for companies not to be the company that encourages somebody to come to harm."

Gluck's closing thought was perhaps the most broadly applicable: "A big mistake would be treating this as an either-or conversation. The harder and more useful question is: what specific features, what type of data use, what privacy protections should we be considering? And whatever rules policymakers do choose - making sure they are clear enough for companies to operationalize and for regulators to enforce."

A genuinely hard question

What makes this debate worth paying attention to – wherever you sit on the policy spectrum – is that the stakes are real in multiple directions. There are documented harms: young people forming attachments to systems that cannot reciprocate, being led toward unhealthy content, or using chatbots as a substitute for human connection and professional support when they need it most. These aren't hypothetical.

But there are also real costs to over-restriction: cutting off a generation from tools that will define the future of research, learning, and work; creating privacy risks through mandatory ID verification; and passing legislation so constitutionally fragile it collapses in court, leaving nothing in its place.

The most honest answer may be that we don't yet have enough evidence to design the perfect policy – and that the goal right now should be building the conditions to get there: better data, better design standards, better-informed parents and better-educated young people who know they're talking to a very sophisticated pattern-matching system, not a friend.

R$