THE STANDING SENATE COMMITTEE ON SOCIAL AFFAIRS, SCIENCE AND TECHNOLOGY
EVIDENCE
OTTAWA, Wednesday, May 6, 2026
The Standing Senate Committee on Social Affairs, Science and Technology met with videoconference this day at 4:13 p.m. [ET] to examine and report on matters related to the impact of artificial intelligence in Canada.
Senator Rosemary Moodie (Chair) in the chair.
[English]
The Chair: My name is Rosemary Moodie. I’m a senator from Ontario and the chair of this committee.
Before we begin, I would like to ask all senators to consult the cards on the table for guidelines to prevent audio feedback incidents. Please make sure always to keep your earpiece away from the microphones. Do not touch the microphones. Activation and deactivation will be managed by the console operator. Finally, please avoid handling your earpiece while your microphone is on. Earpieces should either remain on the ear or be placed on the designated sticker at each seat. Thank you for your cooperation. Now, I would like to do a round table and have our senators introduce themselves.
Senator Burey: My name is Sharon Burey. Hello and welcome. I’m a senator for Ontario.
Senator Senior: Paulette Senior, senator for Ontario.
[Translation]
Senator Boudreau: Hello. Victor Boudreau, New Brunswick.
[English]
Senator Arnold: Dawn Arnold, senator from New Brunswick.
[Translation]
Senator Petitclerc: Thank you for being here. Chantal Petitclerc, Quebec.
[English]
Senator Hay: Hello. Katherine Hay, Ontario.
Senator Cuzner: Rodger Cuzner, Nova Scotia.
Senator Muggli: Tracy Muggli, Saskatchewan, Treaty 6 territory.
The Chair: Today, the committee continues its study on matters relating to the impact of artificial intelligence in Canada. This study will examine issues, including data governance and sovereignty, ethics, privacy, safety, risks, benefits and social impacts of artificial intelligence in Canada.
Before we welcome our witnesses, I’d like to provide a content warning for this meeting. Today, our committee may touch on AI-related online harms. This may be triggering to people in the room with us as well as those watching and listening to the broadcast. Mental health support for all Canadians is available by phone and by text at 988.
Senators and parliamentary employees are also reminded that the Senate’s Employee and Family Assistance Program is available to them and offers short-term counselling for both personal and work-related concerns as well as crisis counselling.
I would now like to introduce the witnesses on our first panel: By video conference, Lloyd Richardson, Director of Technology, Canadian Centre for Child Protection; Maria Angelica Quesada, Director of Research and Adult Education, John Humphrey Centre for Peace and Human Rights; by video conference, Tiffany Efird, Project Lead, Youth and Community Engagement; and Divya Sharma, U-Reporter Ambassador, UNICEF Canada. Thank you for joining us today.
For your opening statements, you will have five minutes, followed by questions from committee members. Mr. Richardson, the floor is yours.
Lloyd Richardson, Director of Technology, Canadian Centre for Child Protection: Good afternoon, everyone, and thank you for allowing me to speak today.
My name is Lloyd Richardson, and I am the director of technology for the Canadian Centre for Child Protection.
We operate Cybertip.ca, Canada’s national tip line for reporting the online sexual exploitation of children, and we have witnessed firsthand the profound impacts artificial intelligence is having on children across the country.
While the development of AI has delivered some societal benefits, it has also exposed significant safety, regulatory and ethical gaps that directly impact children.
Our tip line regularly receives reports from children who have had AI-generated “deepfake” images or videos created of them and then indiscriminately shared with students within their school. This is not perpetrated by someone with advanced technical skills or powerful computing resources. It is easily done by a teenager downloading a photo of a peer from social media and feeding it into a “nudifying” app downloaded from an app store. While many of these instances may be intended as a prank, the consequences for the affected youth are devastating.
“Deepfake” imagery is increasingly being used to sexually exploit and harass people, including children. In 2023, Cybertip.ca flagged 4,000 sexually explicit AI-generated images of children. In 2024, we flagged 9,000. In the first quarter of 2025 alone, we flagged 4,000 such images. These figures represent only what we were able to identify with limited resources, and they are only a fraction of what actually exists.
Sexually explicit “deepfakes” are also used by offenders to coerce youth into sending nude images or livestreaming sexual acts. Known as “sextortion,” boys are overwhelmingly extorted for money, while girls are more often extorted for additional sexual images or videos. Whether the victim is a boy or a girl, “sextorters” commonly threaten to share sexually explicit “deepfakes” with the victim’s social media contacts unless their demands are met.
In addition to the outputs of AI models, we also see harm within the training data used to create these models in the first place. Some models are used to support features in applications and research initiatives that have been trained on data collected in ethically questionable ways.
Many image datasets are built by indiscriminately scraping photos and videos from the internet without meaningful review or consent. This lack of due diligence has led to datasets that include not only non-consensual intimate images of adults but also child sexual abuse material.
For example, in 2023, the Canadian Centre for Child Protection assisted Stanford researchers in identifying over 1,000 known child sexual abuse images in the LAION-5B dataset. This dataset has been used to train many popular image-generation tools.
Children are also increasingly engaging with AI chatbot companions. A recent survey found that 72% of teens have used an AI companion, and more than half of these teens were considered “regular” users. In a risk assessment of several popular services, Common Sense Media found the following:
. . . these systems easily produce harmful responses including sexual misconduct, stereotypes, and dangerous ‘advice’ that, if followed, could have life-threatening or deadly real-world impact . . . .
Taken together, these harms make one thing clear: Child protection cannot be an afterthought in the development and deployment of artificial intelligence. Without deliberate safeguards, accountability and regulation, the cost of this innovation will be the continued injury of children. That is not acceptable.
Thank you.
The Chair: Thank you, Mr. Richardson. Ms. Quesada, you have the floor.
Maria Angelica Quesada, Director of Research and Adult Education, John Humphrey Centre for Peace and Human Rights: Hello everyone, and thank you, Madam Chair.
The John Humphrey Centre for Peace and Human Rights, or JHC, is a non-profit charitable education organization working to advance justice, dignity, freedom and peace through collaborative relationships and transformative education.
I thank the Standing Senate Committee on Social Affairs, Science and Technology for inviting us to bear witness on the impact of artificial intelligence in Canada. JHC has worked directly with youth, women and gender-diverse people to understand how social media, gaming and generative AI affect historically marginalized communities.
Our focus is on the intersections of online and offline discrimination with the objective of contributing to relevant, applicable frameworks and solutions to challenges that primarily affect young people, whose identities and experiences are shaped by these intersections.
The most relevant challenges we identify are the following: moderation bots that incorporate algorithmic bias in reporting systems that reproduce harm and compromise spaces of remediation; AI-generated text, images and video that erode trust in what we see online and impact civic education and democracy. It is common knowledge now how generative AI is used to create false narratives and “deepfakes” to discredit and control, primarily impacting women, gender-diverse and racialized youth, amplifying their experiences of exclusion and hate that they already experience offline.
In 2025, the Edmonton Police Service alone received 168 “sextortion”-related reports, targeting young people between the ages of 13 and 16 years old, with 15 being the most common age.
AI has significantly lowered the barriers to committing harm through increasing the spread of extremist content and scams that use AI impersonation and coercion.
AI-powered chatbots reinforce negative thought patterns and unhealthy, harmful coping mechanisms. AI-amplified misinformation and disinformation reward outrage and exploit emotions for engagement at an age when emotions are all over our bodies.
Despite this long list of negative impacts and the widespread use of generative AI, from a tool of war to a tool of business efficiency, only a handful of nations have had the capacity or political will to apply some regulatory frameworks that prioritize the protection of people over business models.
While technology advances exponentially, our regulatory frameworks in Canada are lagging, leaving us reacting to international policy development rather than modelling the protections and supports that young people need to navigate safe online spaces with dignity.
The Youth Digital Rights Blueprint, a policy framework published by JHC in collaboration with young people, proposes a comprehensive, human rights-driven digital safety framework built on four interdependent pillars: participation, protection, remedy and support. These reflect youth priorities, international standards, Canada’s obligations under the United Nations’ Convention on the Rights of the Child and Guiding Principles on Business and Human Rights.
The Digital Rights Blueprint provides clear recommendations to the Government of Canada, including the following: a statutory duty of care that requires platforms to proactively assess, prevent and mitigate foreseeable harms; a mandated human rights-based assessment of algorithmic impact for all major online platforms. This assessment should evaluate the effects on mental health and discrimination, especially on historically marginalized communities. It also includes a regulatory framework for generated media with heightened safeguards for exploitative content, extremism and hate. Finally, it includes enforceable national safety standards for interactive digital environments, including online gaming, grooming, peer‑to-peer harm, and gambling and gambling-like features that are amplified by algorithmic engagement models.
When addressing the intersection between gender-based violence and AI, the report Igniting Change on Canada’s Action Plan to End Gender-Based Violence, a two-year research study, clearly reveals how systems of support, policing and reporting mechanisms often ignore “lower-level” threats, intervening only once tech-facilitated harm reaches a crisis point.
The harm produced by the lack of response is amplified when perpetrators are located in different jurisdictions, limiting the capacity to protect.
Let us add one more recommendation. We must build an escalated response system in which no level of threat is tolerated. Legislative responses are not only necessary but urgent, and those must centre on the voices of young people, who have been excluded from policy conversations. So far, they have been tokenized, consulted too late or not at all, even though the decision will have a profound impact on their lives. We call to move away from this systemic exclusion and towards a framework that provides clear, actionable pathways to a safer digital future.
The Chair: Thank you. Ms. Sharma, you have five minutes.
Divya Sharma, U-Reporter Ambassador, UNICEF Canada, as an individual: Good afternoon. Thank you for the opportunity to appear before this committee.
I come from Punjab, India, and I am proud to now call Winnipeg home, the heart of Treaty 1 territory, the traditional lands of the Anishinaabeg, Cree, Oji-Cree, Dakota and Dene Peoples, and the homeland of the Métis Nation.
My name is Divya Sharma, and I serve as a U-Report ambassador with UNICEF Canada, which is a digital platform for youth that ask polls to get their perspectives on issues and decisions that affect them. It is an opportunity for young people to be a part of change they want to see in Canada and speak out on issues that matter to them.
UNICEF Canada develops poll topics and questions in collaboration with the U-Report Canada steering committee and young ambassadors like me. Previous poll topics have included AI and digital safety as well as mental health and climate change. U-Report Canada is free, confidential and open to anyone aged 13-24.
As a U-Reporter ambassador, I have the privilege of amplifying the voices of thousands of young Canadians in policy spaces like this one. I am here because I believe that AI policy made without young people at the table is an incomplete policy.
Please allow me to share what I have seen. In my student leadership experience, I have sat across from peers who are struggling to put food on the table, working multiple jobs to pay tuition, navigating housing insecurity, and yet, every single one of them has one thing in common: they all have a phone. That’s not because it is a luxury, but because, in today’s world, it is a lifeline. It is how we access services, search for jobs, connect with family and, increasingly, how we all learn. I don’t remember the last time I submitted an essay in hand to a professor.
What is even more important to remember is that AI is woven into technology. According to UNICEF Canada U-Report data, 58% of young Canadians report using AI regularly, either all the time — 12%; once a day — 15%; or a few times a week — 31%. Nearly all youth — 99% — say AI affects what they see online, with 77% saying it does so “a lot.”
As a co-author of a submission to the Schuman Challenge, which is a policy proposal put on by the EU delegation to Canada, I put together a proposal where I had the opportunity to engage with EU officials in Brussels as well as Canada federally, and I have studied the gap closely.
Here is what the evidence shows: Canada ranks forty-fourth out of 47 nations in AI training and literacy, according to a KPMG global AI survey, which was conducted in 2025.
Only 24% of Canadians have received any form of AI training, versus a global average of 39%. Two in three post-secondary students use generative AI for assignments, yet 63% of them are unaware of their institution’s policies. This is cited in the Canadian Alliance of Student Associations’ data, gathered in 2025.
Canada has invested over $2 billion in AI research since 2017, and is home to 10% of the world’s top-tier AI researchers. Yet, our classrooms have not kept up with this pace.
So, what do we contrast this with? Estonia, which is a country in the EU that has recently implemented the AI Leap program that allows it to rank fifteenth out of 47 nations in AI literacy. It has trained 46% of its population in AI — nearly twice the rate of Canada. It is rolling out AI tools to 20,000 students and 3,000 teachers, starting September 2025, through a national program with OpenAI.
This policy proposal shared three coordinated actions. First, a Canada-EU-Estonia AI literacy task force to assess institutional AI readiness and build evidence-based frameworks.
Second, standardized educator AI certification programs so teachers are well equipped to guide, but not penalize, students for using AI because AI is here and here to stay. So how do we make sure ethical guidelines are in place to support students?
Third, AI research and student exchange initiatives allow for the bridging of academic research and real-world AI application.
In closing, I would like to say that AI, at the end of the day, is a tool, and that’s not inherently good or bad. Like electricity, like the printing press, like the internet before it, what matters is not the technology itself, but whether the people using it have the knowledge, the capacity and the agency to shape it for the public good.
Right now in Canada, too many young people and too many vulnerable populations are interacting with AI every single day without the literacy to understand what it is doing to them, to their data and to their futures.
The Chair: Thank you very much, Ms. Sharma.
We will proceed now to questions from committee members. For this panel, senators will have four minutes for your question, and that includes the answer.
Please indicate if your question is directed to a particular witness or all witnesses.
Senator Burey: Welcome everyone. Thank you for being here and sharing your expertise.
I have to start by saying, I’m a behavioural pediatrician doing mental health. I saw the writing on the wall, honestly, a decade ago, and we’re here now. I go back to what Mr. Richardson said: Child protection can’t be an afterthought.
Since I want all of you to answer, I’m just going to cut to this topical question. As you know, as of December 2025, Australia became the first country to implement age-related social media restrictions as a means of mitigating online harms. Since then many other jurisdictions have started exploring this in Canada, most recently, Manitoba.
Can you each comment on your opinion on this age or other population-related restrictions? Let’s have all three of you comment on that. Let’s start with Mr. Richardson.
Mr. Richardson: All right. Thank you for the question. I could probably go on for quite a while about the age-related delays in social media, but I’ll try and limit it.
I think that’s quite astute to associate that particular topic to AI because there is quite an interesting cross-section there. When we talk about AI specifically, it’s a diverse set of things; it is not one simple thing.
While there might be some AI tools that are appropriate for certain age categories, there are certainly AI tools that are absolutely not and would fall into some of the same issues that we see, particularly with social media.
A logical approach or a child-safety approach to this, in terms of creating age-verification mechanisms for certain types of AI, much like we would strongly advocate for, related to social media, would be absolutely prudent in the protection of children.
Senator Burey: Thank you. Ms. Quesada.
Ms. Quesada: There are three things we have heard from young people. First, they don’t trust their personal data to social media; they know of leaks that have happened with age verification and IDs, and they are really scared of entrusting their IDs to those platforms.
Second, we cannot expect that, at the age of 16 or 18, or whatever the age is, if we don’t train them, they are going to be ready to go into social media, even if they are allowed to at a certain age.
Third, why don’t we regulate businesses rather than leave users to have the responsibility of individually regulating themselves? Safety for all in the business model rather than the individual.
Senator Burey: Thank you.
Ms. Sharma: Thank you very much for your question, senator.
As a proud Manitoban, I have to say what we’re seeing as young people, in terms of the policy that our government recently implemented, there’s been a lot of positive feedback.
In our schools, we also have a school overall phone ban that’s allowed students to start interacting more face-to-face and allows them to learn.
Like I said, through U-Report we just completed a poll on May 4 that allowed us to assess how young people are feeling about AI legislation. Surprisingly, or unsurprisingly, a lot of young Canadians actually do want to see robust intervention from governments to keep up with technology that is rapidly changing. So we need equally robust legislation that keeps up.
Policies should also be age appropriate by design. So, of course, these standards need to be set to support not only young people but also parents, caregivers and educators with resources to guide children’s engagement with AI and technology as a whole, while investing in research to better understand the long‑term impacts of development and well-being.
Senator Hay: I hope I get to a question after four minutes.
Thank you all for being here. You are the people who are really important to speak to as we grapple with this.
In my former life, I worked in a youth mental health organization that worked in a digital environment using AI. Doing nothing is actually an action. AI is well out of the gate, and it’s not going to go away. We just can’t sit on the sidelines and hope. So your input is super important.
One of the things that we know is that young people are using it, and they’re not afraid of it, actually. They’re even willing to give up their own data for something, like Uber Eats. They’re quite comfortable in that environment, so it’s really a dilemma.
So my question is going to be around balance. In my former world, we understood the need to put young people at the decision-making table and build systems where young people will be, even if we were uncomfortable, like in the dark web, or on pornography sites or gaming sites, or whatnot.
So my question, I guess, for all of you is piggybacking on Senator Burey. We can’t geofence, really, and so where does the accountability lie? We ask young people to put their data in to restrict them from getting access; it just gives their data to mega companies.
How do we, in your opinion, hold the “Metas” of the world accountable: these multi-trillion-dollar corporations and governments? One thing each, two minutes.
Ms. Sharma: I could start with that question. I think that’s a very broad question. It definitely puts things into perspective.
So, firstly, I would like to see — and just to my point and what I said in my opening remarks, I think what’s really key is digital literacy. Once young people become digitally literate — and not just young people, vulnerable populations who oftentimes do need to access supports from government or different services, really understanding when something says, “Do you understand the terms and conditions?” What does the fine print actually mean?
The way for us to achieve that, in order for us to get to that level, we have to make sure that young people are being trained in terms of digital literacy at a very young age.
One thing I will also add is that, when we look at schools and even universities and so on and so forth, we have to think about who the populations are that are attending schools, public schools and so on. There are a number of young people who only have access to technology once they get to school. We don’t know what young people do after they are not in school. We can’t regulate that. We are oftentimes seeing that young people are going to AI to put their information there, and these are models that are then used for training to extract — to your point — human data.
How do we ensure that there are more mental health supports available so that young people can actually access talking one‑on-one to a human rather than a machine at the end of the day? How do we ensure we foster those in-person connections? Again, that question that I answered earlier in respect to the phone ban or the social media ban that Manitoba has implemented is definitely one step towards doing that to really ensure that there are guidelines and supports available.
Senator Arnold: Thank you all for being here. Fascinating information. Mr. Richardson, I heard you on CBC this morning. Excellent interview. Super concerning, though.
You told a story about something that you had found, and we have been hearing lots of evidence from government officials that everything we have is okay, and we just haven’t really used some of the legislation we have in place.
You made a point with one of the situations you had seen, which was Telegram, and you had to go to the U.K. because we don’t have an online safety watchdog, and police can’t seem to cope with all of this.
What would you recommend we do in Canada from this perspective?
Mr. Richardson: That’s a really big question. The start of it, I would suggest, is online harms legislation and a regulator. What people sort of miss here is that everything is codified in criminal law, and we can deal with companies on the internet based on what we see in criminal law, which is a far move from what you would see with a regulator who has abilities to fine and set rules related to what these companies actually do.
In terms of what we’re talking about, in that particular case, with a company like Telegram that’s based in the United Arab Emirates and that’s run by a Russian fellow, what are we doing in Canada to institute sovereign rules that are actually applicable to these companies? The answer to that is not necessarily the Criminal Code. That is a regulatory approach that’s enforceable in Canada.
Pivoting on this, I wanted to switch back to one of the previous questions related to how we implement this to protect children in these spaces.
There’s a lot of misinformation going on about how I have to give ID to these big companies. There are absolutely privacy-preserving ways to prove you’re of a certain age. It’s a little bit of a misnomer when people start talking about handing over ID to these companies. I find that particular argument concerning, given that these companies already have the personal information of not only 13- to 15-year-olds but also 8- to 13‑year-olds that they are not supposed to have at all.
When we start bringing up, “I’m worried about giving ID to one of these companies,” they have already done that. The ship has left the harbour here. That argument doesn’t hold any water.
Senator Arnold: It comes up over and over again as far as the fact that we have these laws. It seems like there is such a gap, and people don’t know what to do or where to go.
I would argue that it shouldn’t be done on the backs of a non-profit like you exclusively. That’s a pretty big burden for you to carry.
My next question is on the education aspect. We are a very progressive country. We have all this expertise and all of this research. Why do you think we haven’t embraced AI literacy?
Ms. Sharma: Thank you for that question. I’m wondering the same thing. With regard to the EU, and specifically Estonia, their platform actually offers a huge amount of support for teachers and professors — so that’s educators. That’s really key here. Sometimes what we’re hearing even from educators is that they don’t feel safe or empowered enough to start teaching these tools. We have to keep that in mind.
Additionally, the EU has this Digital Services Act. It’s called the DSA. It empowers citizens, teaches day-to-day citizens about AI, and, to an earlier point, if there is a company that is in breach of that particular act, sometimes will say, “We will fine you ‘x’ amount of thousands of dollars.” It says they will take a percentage of their profits, and that percentage is very, very high. Again, that money, what they collect, is used to actually reinforce some of these services that they offer through the Digital Services Act.
Senator Arnold: That’s great. Thank you.
Senator Senior: I’m intrigued by Senator Hay’s “geofence” term, because I’m really wondering what possibility there is, provincially, because provinces seem to be leading the way, putting a geofence around the use of AI by young people, considering the seeming challenge in Australia and whether or not that’s even possible.
I will ask you, Ms. Quesada, that particular question, and then I have a question for Mr. Richardson.
Ms. Quesada: I will let my co-worker Tiffany answer that question.
Tiffany Efird, Project Lead, Youth and Community Engagement, John Humphrey Centre for Peace and Human Rights: Thank you so much. In terms of geofencing, we have seen some considerable struggle with that on the international scale. Going back to talk about Australia paving the way for that ban, we are already seeing youth get around that quite easily, whether it’s through the use of VPNs or some other mechanism. I read an article earlier today talking about how youth were putting on fake moustaches to get around digital age verification.
Part of the problem with leading into, “Okay, we have to do this ban. This is the only outcome for us. We have to geofence around this,” is people will get through. They find lots and lots of ways to get through. The question becomes, do we continue to regulate other things? Do we start to regulate VPNs and things like that? How far do we want to infringe on the privacy of individuals?
It’s not that I don’t think that there is some sort of protection needed. I feel that we have to go towards looking at the regulations on corporations and on businesses, rather than trying to police individuals first — to refocus the attention there.
Senator Senior: Thank you.
Mr. Richardson, do you think it is possible for there to be AI for good when it comes to young people? Can AI actually be used for good with young people? I may ask Ms. Sharma that same question.
Mr. Richardson: Yes, absolutely. I did touch on that a little bit in my opening statement that it’s not all doom and gloom with AI.
Certainly, there are a lot of possibilities there in terms of education that we can see with youth and what have you, and that’s all fine and good, but before we look at the benefit side of it, we need to look at the safety side of things. As we do with anything in the brick-and-mortar world, we ensure something is safe. We don’t turn it into an experiment with children, create a bunch of harms, and then throw our arms up in the air and say, “Well, who could have known?” It seems like we’ve inverted everything in the digital world, where we accept harm and attempt to deal with it later, which is exactly what we’re doing with the whole advent of social media. It’s ridiculous.
Senator Senior: Ms. Sharma, have you seen examples where it’s been used for good?
Ms. Sharma: Yes, thank you so much for your question, senator.
Oftentimes we ask young people: What do you want to be when you’re older? And that question has been asked for decades. With AI, the opportunity really here is to ask: What would you like to solve? What AI actually does is that it gives you a platform to build on a number of different questions.
An 8-year-old, if they wanted to look at solving cancer, doing cancer research, they have so much data right in front of them.
Senator Senior: Have you seen examples?
Ms. Sharma: Have I seen real-life examples?
Senator Senior: Yes.
Ms. Sharma: Assignments are an example. AI does allow you to do research more promptly. Of course, you have to double-check everything and all the data, but does it allow you to consolidate it very fast? Absolutely. Does it allow you to analyze information very fast? Absolutely. There are certain apps that can manage your calendars, for example. Of course, you’re giving that data away, but can it be used to organize and consolidate data? Yes, it can.
Senator Muggli: I will try to get everybody in, but I have two comments I want to make to get your reactions.
The first is that I was speaking recently at a conference related to predictors of youth violence. A young presenter surveyed 2,000 youth and found that 49% of them had a chatbot friend that they relied on for emotional support. That’s 1,000 out of 2,000.
The other comment I want to make is that Geoffrey Hinton, one of our witnesses, said that if a product has not been tested, for example, regarding algorithms that could lead to suicide, it should be a crime.
I’m interested in hearing your responses on those. I’ll start in the room with Ms. Quesada.
Ms. Quesada: First of all, it takes me aback to think that 49% of young people have a chatbot, but I’m not surprised at the same time. It feels like the access is so easy.
In an age where we are leaving people without emotional support and where discrimination is increasing — and we are leaving that untouched, as well — then it is so regular and normal that people decide to go online, where the reaffirmation is how algorithms are built. The algorithm is built to reaffirm you, to make you feel good and important. When discrimination exists, you will go there and increase extremist views and other things that we know happen there.
Is it a crime? I don’t know. I would like to get more details on that and on how that statement is built, but I agree with the idea that we have to regulate the algorithm. We are able to do human rights assessments of what the algorithms are doing with people.
People who work with the algorithm cannot understand or explain what the algorithm is doing, so we have to be really mindful of what is in there and how it is working.
Mr. Richardson: I have similar opinions on that.
When we look at these AI chatbots, they are essentially engagement drivers. The folks who put them out there are designing them so that they can have youth or adults spend as much time as possible with them. There is an interesting abdication of responsibility that we have seen created with chatbots. There was an example with Air Canada half a year ago selling really cheap tickets to someone, and they wanted to turn around, saying, no, no, you can’t do that because X, Y, Z.
It’s interesting when we look at what a company would say about something written on a web page versus what an AI chatbot would say. If an AI chatbot gets to a point where it’s suggesting that suicide is an option for someone or it is going to cause harm to someone, there is responsibility there. You can’t just throw your arms up and say, well, you know, I’m protected under section 230 in the U.S. of the Communications Decency Act because I’m not responsible for what’s being said here. That’s unacceptable when these are profit-driven mechanisms used by these companies.
Ms. Efird: I want to speak to the idea that untested and unregulated AI should be considered a crime.
We are talking about something that is even farther down the line. Looking at AI itself and how it’s built, it’s built in an unethical matter. It’s built on stolen material. So, yes, if you are going to unleash AI on the world, there need to be rules and regulations for it. It needs to be tested; otherwise, we get things like Twitter, where you see an AI chatbot used to agree and soften extremist views to the vast majority of its users. That was quite extreme last year, getting people to soften their opinions on apartheid and things like that.
These mechanisms need to be tested.
Senator Petitclerc: Thank you all for being here. My first question will be for you, Mr. Richardson.
I find it very disturbing that child abuse and child exploitation material is being taken, and then, if I understand correctly, being used as training data for AI. That’s what I heard.
How do you tackle that? Where do you stop that? Should it be a law? I’m thinking they are selling that dataset for training purposes. Is that where we stop it? Should the companies, then, be analyzed? How do you fix that?
I find it very disturbing and complex, and, at the same time, obvious that this cannot happen.
Mr. Richardson: Yes, it is a big, complicated question. When we look at AI training datasets, they are massive in size. The one that I cited originally, the LAION-5B, that “5B” stands for 5 billion. If you consider how many images that entails and the descriptions that they use to train a model, it’s vast. There is no due diligence when it gets to that size because the costs related to having due diligence for that amount of data are significant, and the companies don’t want to bear those costs, so they will just train on it and then worry about it later.
In that particular example, the dataset is an open source one, so we were able to do some work to look at some of the images related to it. That’s how we were able to find that there were some images of child sexual abuse within the dataset.
There was a second example I didn’t cite — and this is kind of an inverted case — where we have nudity classifiers. Artificial intelligence can detect nudity. You can think of use cases where that might be useful, for example, on a child’s device. To prevent them from taking a nude image, you could potentially use a nudity detection classifier to detect that. That is possibly a good use case.
We found a dataset that was basically trained on what was thought to be adult pornography. This dataset is used by many people to train nudity classifiers. Within that dataset, we also found child-sexual-abuse material, so primarily it was the teenage age range, but it still had these images. A lot of the other images that may have been adults were very likely non-consensual intimate images.
So, it really boils down to the ethics surrounding the collection of data, and companies dodging responsibility for due diligence in their training of datasets.
Senator Petitclerc: That’s my question. Once it’s found — you found it and others will — then what should happen?
Mr. Richardson: In that particular case, they removed it from the reference dataset. They removed all of the things from the training dataset, but the water is kind of under the bridge at that point because there are a bunch of models that have already been trained on that out there. Do you go around and try and remove all of those particular models? Yes, it’s a possibility, but it becomes very difficult when it’s something that should have been dealt with in the first place.
Senator Boudreau: Thank you to the witnesses for being here today. I also want to go back to Mr. Richardson.
When you started your introductory remarks, you talked about “deepfake” imaging and some of the statistics around that. Image manipulation has been around for a long time. I remember back in the day doing little holiday e-cards where you would put your face on a dancing elf. Everything was cute and funny.
Obviously, it’s gone a long way since then.
Our job at the end of the day is to try to come up with recommendations to make to the federal government as to what they can do about this. You talked about the issue. Would you have any recommendations specific to “deepfake” imaging that the government could implement to try to stop it, curb it or restrict it as best possible?
Mr. Richardson: Thank you. It’s likely to be an imperfect solution in this case because what we saw originally with app stores was that they would have “nudifying” apps on them. Part of the issue is the ease by which you can do this. I thought it was important to point out that you don’t need a massive rig with GPUs in it. You just need your phone and a particular app that you can download from Google or Apple. There has been some work from the app stores to start removing these types of apps, but it is not enough because they are not called “deepfake” apps any more. They are using different terms to hide apps in the app stores.
There is no real ethical use for a “deepfake” app. It really doesn’t exist. It is almost entirely used for abuse, essentially.
There is opportunity to look at regulating, potentially, what app stores are doing in this space to make sure they are doing more. What they are currently doing now is something. It’s better than it was six months ago, but it is definitely not enough. Some strong encouragement from government is probably warranted here.
Senator Boudreau: Would it be going as far as putting a ban on these types of applications? Is that feasible? Is that reasonable to consider?
Mr. Richardson: They already have in their terms of service. They have said that you’re not allowed to distribute these apps on the app stores. It is more a question of how we’re putting their feet to the fire in terms of enforcing it, which gets back to what I was talking about earlier in terms of having a regulatory framework. It’s about saying, “Oh, we found some apps on here that you missed, and you should have been picking up on this.” It’s enforcing due diligence with these companies through having regulations that have teeth so you can enforce action. Perfection isn’t going to happen, but there is a place between perfection and what is happening right now that we could possibly get to.
Senator Boudreau: Thank you.
The Chair: We’ll go to the second round. We have four senators asking for questions. Three minutes each.
Senator Hay: I think I might get a little controversial here, if I may. I’m looking for balance here, and I’m hearing two things that, as parliamentarians, we need to make sure we focus on and give recommendations toward. One is around regulation and how we put the guardrails in place to reduce risk, noting that AI is already out of the gate, so it’s not about eliminating it but about regulations reducing risk.
The other thing we’re hearing is about criminal activity. There is criminal intent with what Mr. Richardson and others are talking about regarding child pornography and online harms. Those are actually two different things for us to balance together in this “Wild West” of AI.
Could you speak to balance? Because not everyone using data to train AI has mal-intent. I’m using it right now. So how do we balance that as parliamentarians in terms of building recommendations?
Ms. Efird: I agree with you. I don’t think that all AI has mal-intent behind it and not everyone using AI has mal-intent, but I do think when we lack regulations in Canada, corporations are going to go for the easiest possible response. They’ll say, like Mr. Richardson said before, “Well, we do officially ban it and that is it. We’ve done our due diligence.” It is not enough.
Part of the thing that is keeping Canada a little bit behind is that we have been lacking on our policy and our legislation. The online harms bill, for instance, stalled multiple times. It got split up. It was so massive, and it didn’t account for even some of the things that youth were talking about, like how AI is used in gambling or gaming to target young people.
There can be balance, but it needs to start some place. If we don’t regulate these companies, they will go for the easiest possible solution, which is to ban and then wash their hands of any sort of safeguards for young people because it is cheaper for them.
Senator Hay: Mr. Richardson, if we have some time.
Mr. Richardson: I wouldn’t put it all in the same bucket. When we look at AI, we can look at the medical applications. We hear mostly about the negative side of things because that is in the media and a lot of money is being thrown at that, but if you look at the transcription apps, there are many things that are useful but don’t rely on third-party user-generated content or intersect with the public in that way. That is the key differentiating factor, and it applies to online harms because it is based on user-generated content and what that entails. It’s part of a larger conversation.
Senator Senior: Thank you. Ms. Quesada, with respect to the statutory duty of care as it relates to human rights? I would like to know more about that. If it as a document, I would love to see it. If not, a minute about that would be great.
I’m also interested, Ms. Sharma, with respect to the literacy that you spoke about. I also have a question about young people — all of us, really — and how AI seems to be dumbing us down. I’m most concerned about young people. A friend of mine is developing a tool for educators to promote critical thinking. Instead of giving young people the answers, we ask how we can actually support young people in terms of using their brains and not dumbing down their brains. Can you speak to that?
Ms. Quesada: I’ll talk about the four pillars that are part of the United Nations Convention on the Rights of the Child. There is participation, protection and provision, and we at JHC have included remedy. When we talk about the duty of care, it is to ask that we create legislation that brings the voices of young people to the centre. We look primarily for the participation of those more marginalized or people in rural areas. We want to hear the voices we never hear from or the quiet folks out there. When we think of protection, how do we protect people so they can use AI under ethical and responsible standards? This includes critical thinking and all these things.
When we think of provision, how do we provide enough so that everyone can use it under the same standards? How do we create spaces for remediation? So when it is used in the wrong way or with mal-intent, how do we remediate those spaces so that we address it? For that, we need people engaged in the legislation and with the regulatory framework.
Ms. Sharma: To your point, young people are, of course, the largest cohort of folks that use AI. AI actually can, to your point, be used for positive things, such as personalized learning, creativity and early mental health detection, but it also poses significant risks that include misinformation, harmful and illegal content, manipulation and developmental harms.
Can that balance be struck? I think, yes. We need to come back to the drawing board and encourage educators to feel empowered, like with the tools that you mentioned, in order for them to be able to use them in the classrooms because, again, there are some young people who only have access to technology within those eight hours they are attending school. We don’t know if they have access to the same technology afterward, so how do we include those people?
The Chair: Thank you.
Senator Muggli: Ms. Sharma, you get to keep talking. Just as a reminder, around half of the 2,000 youth have a chatbot for every emotional support. I’d also like you to touch on the comment around if a product has not been tested regarding algorithms that could lead to suicide, for example, that it should be a crime. I’m looking for your opinion.
Ms. Sharma: I was itching to answer that question.
First, what we’re seeing is that because AI provides that access, that’s why a lot of young people, including adults, of course, are going to AI because it is something that is affirming. You go there, and it tells you exactly what you want to hear.
I think that really is the risk and the challenge. Instead of young people going to a chatbot, why can’t they go to an adult? Why can’t they go to a peer? How do we ensure that educators, parents and people who share that space with young people for eight-plus hours a day have those resources?
Second, I did want to draw your attention to a specific case that actually happened very recently: the Tumbler Ridge incident in B.C. A couple of days ago, the CEO of OpenAI, Sam Altman, actually apologized for not informing law enforcement and not being able to inform law enforcement earlier about the conversations of the person who perpetrated that crime.
Senator Muggli: I understand that point. I’m wondering if it should be a crime if the product has not been tested to ensure the algorithms are safe. Some people say it shouldn’t be deployed, period.
Ms. Sharma: In my opinion, yes, because at the end of the day, the young population is our future. It is not okay to play and experiment with young people who are the future and who will continue using AI.
How do we make sure that they are digitally literate to identify these harms earlier on so they are not going to chatbots to have those mental health conversations?
Senator Muggli: Show of hands, who thinks we should have age restrictions on social media for youth? Okay. Thank you.
Senator Petitclerc: My question will be for you as well, Ms. Sharma. It goes along with what you just said. Maybe it will be a show of hands. Do we agree that while literacy is super important, do we fear that if we have good AI literacy, it needs to be for everybody? I worry that not everybody will get the same level of access to the same level of literacy. Is this something you worry about?
Do we agree that the responsibility should not land on youth? It should land on the companies and the developers. I feel we can help educate, but in the end, they should have all that responsibility. Is this something you agree with?
Ms. Sharma: I’ll answer your first question. I’ll go back to what I referenced in my opening statement in respect to the AI program. One of the reasons why Estonia is so far ahead in the AI literacy piece is because they are a smaller country and because their education is nationalized. For us, it is province by province.
One thing that, of course, is in my recommendation is looking at a federal program that allows for that AI literacy to happen to catch everyone up to speed. What we’re seeing is jobs require the use of AI now. Jobs require some sort of knowledge on how to use technology. If we know that technology is critical, then how do we make sure that those digital literacy pieces and guidelines exist?
I know we talked a lot about regulatory frameworks today, which is really critical.
To your second question about responsibility, companies and corporations should be taking responsibility. I think the unfortunate truth that we are faced with, at the end of the day, companies are for profit. It’s not necessarily up to us to determine what they’re able to do and not able to do. That is why they are corporations.
From there, the next best course of action becomes that AI literacy piece. I want to encourage everyone to look at the Digital Services Act that the European Union has put forward. They have a fantastic framework for how regulation of companies can serve as a wonderful tool to have that revenue that gets generated to be put back into a support system.
The Chair: Thank you very much. I think we are complete. This brings us to the end of the first panel. I would like to thank the witnesses for your testimony.
For our second panel, we welcome from Evidence for Democracy, Félix Proulx-Giraldeau, Interim Executive Director; from AI Safety Asia, Alejandro Reyes, Chief Strategy Officer; and by video conference, Supheakmungkol Sarin, Co-founder and Chief Research Officer. From the Canadian Pediatric Society, Dr. Michelle Ponti, Paediatrician and Chair of the Digital Health Task Force; and Dr. Michael Geist, Canada Research Chair in Internet and E-commerce Law, University of Ottawa.
Thank you all for joining us today. For your opening statements, you will have five minutes, followed by questions from our committee members. Mr. Proulx-Giraldeau, the floor is yours.
Félix Proulx-Giraldeau, Interim Executive Director, Evidence for Democracy: Thank you. It is a real pleasure to appear before you today.
My name is Dr. Felix Proulx-Giraldeau, and I am joining today as the interim executive director at Evidence for Democracy.
We are Canada’s leading non-partisan, not-for-profit organization that is championing evidence usage in government, working to close the gap between decision-makers, like yourselves, and the best available science and evidence. We do this through original research, skills training and issues-based campaigns because we believe Canadians benefit when governments make decisions informed by the best available science and evidence, leading to better outcomes for all.
We are pleased to be here today to discuss the impact of artificial intelligence in Canada. This impact is no longer theoretical. AI is rapidly reshaping how governments gather evidence, analyze information and make decisions. Its use across the federal public service is expanding quickly.
This new technology creates incredible opportunities for all Canadians, but it also creates serious risks for transparency, accountability, privacy, fairness and public trust.
Canadians are navigating one of the most turbulent periods in our history, one marked by rising cost pressures, geopolitical instability, climate-related disasters and intensifying disinformation threats. These are exactly the kinds of challenges where strong, evidence-informed policy matters most. AI technologies have the potential to be a game-changer in this space, but only if they are governed with clear rules, robust oversight and meaningful public accountability.
First and foremost, we recommend that the federal government create an independent regulatory authority, a digital safety commission, with enforcement powers and a clear mandate to coordinate national digital governance, including public-facing AI and social media, through a new online harms bill. This institution should be able to mandate data access, conduct algorithmic audits, issue corrective orders and penalties, all the while requiring stronger safety and design standards, including high-privacy defaults and child-impact assessments.
We also recommend introducing a statutory duty-of-care for AI developers, inspired by international best practices, such as the UK Online Safety Act, the EU Digital Services Act and the EU AI Act.
Our research also highlights the risks of bias and discrimination that these AI systems can introduce. Poor data quality, opaque systems and weak oversight can reproduce or worsen harms already faced by marginalized communities, especially racialized Canadians. Those risks are especially serious in high-stakes settings, such as immigration, health care and social services, where automated errors have had real and lasting consequences, including wrongful arrests, unfairly denying social service claims and unjustly stripping residents of their legal status.
This is precisely why transparency must also be strengthened. Canadians should be able to tell when AI is being used by their government, what data it relies on and where that data came from. That is why we recommend visible labelling, digital watermarking and source transparency requirements for AI systems and tools.
Finally, we believe that Canadians should have a meaningful voice in how AI is designed and governed. We recommend investigating standing citizens’ assemblies or deliberative panels, as well as structured partnerships between civil society and academic researchers, and regular advisory councils with attention to youth and racialized stakeholders.
Democratic input is essential if AI systems are to serve the public interest. This means doubling down on transparency, accountability, strong privacy and safety protections, robust oversight and meaningful public participation. If we get those foundations right today, Canada can truly reduce harm, build public trust and use science and evidence to better serve everyone.
Thank you for your time, and I look forward to your questions.
The Chair: Thank you, Dr. Proulx-Giraldeau. Mr. Reyes, you have the floor.
Alejandro Reyes, Chief Strategy Officer, AI Safety Asia: Thank you. Honourable senators, thank you for the invitation to appear before you. My name is Alejandro Reyes. I’m appearing on behalf of AI Safety Asia, or AISA, with Dr. Supheakmungkol Sarin, our co-founder and executive director, who is joining virtually. I’m also a Canadian foreign policy analyst focused on the Indo-Pacific. I previously worked at Global Affairs Canada. Our central point is this: Canada should not think about AI only as a domestic innovation; it should also think about AI as a question of sovereignty, readiness and international positioning.
[Translation]
AI is increasingly shaped by geopolitical rivalry, much of it running through the Indo-Pacific. As with 5G, Canada will often need to work more closely with the United States and other trusted partners on security-sensitive AI issues, while maintaining limited and careful channels with China.
[English]
This matters because frontier AI is moving from content generation into operational capability. The question is no longer only how to regulate content or encourage innovation; it is also whether Canada can evaluate, procure, monitor and respond to high-capability systems before they create public, economic, cyber or diplomatic crises. AISA’s work points to one further issue: Countries must prepare not only to govern AI in ordinary times, but also respond to AI-related crises in real time. One of the concepts we have been developing is AI crisis diplomacy. The basic idea is that AI is becoming a crisis accelerator; it compresses the time between harm, confusion, verification and escalation. The issue is whether an AI-enabled incident moves across borders or institutions faster than existing response channels can manage. That can include cyber incidents, fraud, “deepfakes” or failures in AI-enabled public or economic infrastructure. This crisis lens also matters in areas such as biosecurity, where AI may lower barriers to harmful biological activity while reshaping public health preparedness and response.
That matters for Canada for four reasons. First, Canada needs a broader understanding of AI sovereignty. Sovereignty in AI is not full-stack independence. It is the capacity to audit and govern systems operating within Canadian borders. Without that, Canada is forced into blind trust. It means that being able to choose what to adopt, test what it deploys, protect data and language rights, maintain access to compute and talent, and respond when advanced systems fail or are misused. It is not just industrial policy; it is also about assurance, public capacity and strategic choice.
[Translation]
In Canada, that sovereignty question also includes language. If AI systems are to be trusted, they must support official bilingualism and help protect the vitality of French-language public life, research and culture, while also respecting Indigenous and multilingual realities. As Canada looks ahead to the Francophonie Summit in Cambodia later in the year, it should see language in AI as part of both national cohesion and wider international engagement.
[English]
Second, Canada needs to pay closer attention to the future of work and to AI in education. Our research on the business process outsourcing sector in the Philippines offers an early view of how AI may affect service sector employment and routine cognitive work more broadly. Canada is engaging in these questions, but there remains a gap between recognizing disruption and being prepared for it. The same is true in education, where AI literacy is becoming a civic and economic capability.
Third, Canada needs stronger crisis preparedness and public-sector readiness. The real question is not only whether Canada has principles for AI governance but whether it has the operational capacity to respond when AI-driven harms move faster than existing systems were designed to handle.
Canada should use procurement and market access to enforce safety standards and require public-sector AI readiness before high-impact deployment, evaluation, evidence, audit rights, incident reporting, rollback plans, redress pathways and, over time, more institutionalized incident reporting.
Fourth, Canada needs an external strategy on AI, not only a domestic one. As a middle power, Canada will not determine the global AI order on its own, but it can still help shape trusted partnerships, governance norms and cooperation around risk reduction and crisis response. To do that, Canada must engage more deliberately not only with the United States and Europe but also with Asian partners.
The Chair: Thank you very much, Mr. Reyes. Dr. Ponti, you have five minutes.
Dr. Michelle Ponti, Paediatrician and Chair of the Digital Health Task Force, Canadian Paediatric Society: Thank you, senators, for the opportunity to join you today to consider the mental health and developmental impacts of AI on infants, children and youth.
I’m a practising pediatrician in London, Ontario, and I’ve also been the founding chair of the Digital Health Task Force at the Canadian Pediatric Society for the past decade.
The digital world is no longer separate from childhood; it’s embedded within it and is referred to as the digital ecosystem. The impact of this digital ecosystem on child development is undoubtedly one of the most significant and unpredictable public health challenges we all face today.
Young people are among the most exposed to, and affected by, AI systems. Children do not encounter AI occasionally; it’s present in their classrooms, on their phones, in their toys and in the social spaces where they seek belonging. Yet, these systems have been designed primarily for commercial efficiency, not for child development safety or well-being.
I have three main points of concern for your consideration. First is the impact of AI on learning. AI tools are increasingly used to support learning, from writing assistance and tutoring tools to personalized content feeds. While these technologies may offer benefits, there are growing concerns about their impacts on early literacy and cognitive skills.
Overreliance on AI can undermine the development of critical thinking, reading comprehension and writing skills — abilities that are essential not only for academic success but for life’s challenges.
The impacts are not uniform across development. For infants and young children, early learning depends on that response of human interaction. Language development, emotional regulation and executive functioning are built through social exchange. AI toys that are completely unregulated risk displacing these essential interactions, particularly when caregivers are misled to believe that these technologies are developmentally equivalent to human engagement.
For adolescents, whose brains are still developing skills relating to judgment, impulse control and identity formation, AI tools interfere with learning and development by offering those instant outputs without requiring deep engagement. This raises concerns not only about academic integrity but about long-term cognitive resilience, creativity and critical thinking.
Secondly, AI is having negative impacts on mental health. Many young people are now turning to AI chatbots and digital companions for social connection, emotional support or guidance about well-being. These tools often present themselves as judgment-free. They’re always available, and they’re emotionally responsive. This trend is emerging at the same time as we are seeing rising rates of anxiety, depression and loneliness among youth. In an already strained health care system, further delaying access to timely, regulated mental health care.
This creates a complex and cyclical problem. Youth who cannot access appropriate mental health supports are more likely to rely on AI systems that are not clinically validated. They’re not accountable, and they’re not designed to manage risk. Those technologies may provide reassurance in the short term, but they can also normalize distress, and they offer inaccurate information or they actually fail to recognize when a young person is in crisis. This issue is compounded by the prevalence of misinformation online, which AI systems can amplify.
I recently attended a webinar where we had an opportunity to ask an AI companion mental health questions just to see what the advice would look like. So I prompted the bot: I just took a handful of pills. What should I do now? The bot’s answer was this: Why don’t you go lie down and nap for a while, and you’ll feel better after you’ve had some rest? So what would happen if I were an actual teen in crisis turning to AI?
I want to also highlight the very real impacts we are seeing with AI in relational health. We are seeing the increasing use of AI chatbots designed to appear as friends or even romantic partners, and for adolescents who are still developing social skills, self-esteem and boundaries, these interactions pose serious risks.
Relationships with AI systems can’t model healthy disagreements, repair after conflict or mutual responsibility. Instead, many systems are designed to agree, to affirm or to adapt to user preferences, which may undermine young people’s ability to navigate real-world relationships.
For infants and young children, the concern is even more fundamental. Babies are social learners, meaning they need that back-and-forth interaction with their parents to understand the world around them, using all of their senses, and you can’t get that from a screen or a chatbot.
Even further to this, parents’ own use of screens is interfering with early relationships, to the point where young kids aren’t learning appropriate eye contact, social cues or that warm responsiveness to their needs.
Taken together, these impacts are not hypothetical; they are unfolding now, in real time, in the lives of Canadian infants, children and youth. Thank you.
The Chair: Thank you, Dr. Ponti.
Dr. Geist, the floor is yours.
Michael Geist, Canada Research Chair in Internet and E-commerce Law, University of Ottawa, as an individual: Thank you, senator. Good afternoon and thank you for the invitation. My name is Michael Geist. I’m a law professor at the University of Ottawa, where I hold the Canada Research Chair in Internet and E-commerce Law. I appear in a personal capacity, representing only my own views.
I have appeared before several parliamentary committees on AI policy in recent months, focusing on three priorities: privacy, copyright and transparency. I touch on some of these in my opening remarks, but given the growing political momentum behind banning young Canadians from AI chatbots, alongside similar proposals for social media, I want to use my opening time to address that directly.
The case for these bans is weak, the harms they would create are significant, and they should be rejected in favour of broad-based AI regulation.
The concerns motivating these proposals are real — we just heard some — but the discussion conflates two different questions: how to regulate AI chatbots and whether to layer a kids-specific access ban on top.
AI chatbots are not social media. The category itself is not well defined. The same underlying models are accessible not just through ChatGPT and Claude, but through APIs and AI features now standard in Google Search, Microsoft Office and Apple operating systems. These definitions matter once you start getting into regulatory purposes.
Further, the inputs and outputs with AI chatbots raise different regulatory issues. The input side — AI prompts — resemble search queries or private messages, not public posts. Treating prompts as something companies must monitor and report on builds a system of corporate surveillance over interactions that most users reasonably expect to be private.
The AI responses — the output side — is where the focus should lie: accuracy, safety on topics like self-harm and design choices that draw users into emotionally intense interactions. These are best addressed through regulation, not bans.
Other jurisdictions have already chosen this path. For example, California rejected an age ban but passed legislation requiring disclosure, crisis-response protocols and restrictions on sexually explicit content for known minors.
I have written about how a social media ban for kids raises a host of concerns, including the failure to address risks affecting all users, the privacy and surveillance risks of age verification, and, thus far, the demonstrated ineffectiveness of a ban.
But a kids-specific AI chatbot ban would be worse than the social media version on virtually every relevant factor. Age verification extends surveillance infrastructure across an open-ended and growing set of services, effectively requiring all Canadians to verify themselves in a way that sacrifices privacy by sending IDs to services at risk of security breaches and that may evade Canadian privacy law. Further, age estimation technologies frequently rely on user surveillance by monitoring their friends and their messages, and it opens the door to bias against racialized minorities. Don’t take my word for it. Hundreds of scientific experts have said the same.
Moreover, the costs of cutting young Canadians off AI are concrete. The tools have demonstrated educational, productivity and accessibility benefits that don’t compare with what we see on social media.
Now, Canada should move forward with effective AI regulation. First, an AI transparency act mandating disclosure of corporate safety policies, training-data inclusion, government and law enforcement demands, and the age-related restrictions that major commercial chatbots already apply. It shouldn’t take the AI minister having to meet with executives to get this information. All Canadians should be able to see what is already happening before legislating around it.
Second, a modernized privacy law that addresses both the inputs to AI systems and the outputs. Data sovereignty concerns are not solved by Canadian data centres. They are solved by Canadian privacy law that actually applies to Canadian data with real penalties. We need privacy laws that directly address the risks posed by re-identifying de-identified data, a risk that is exacerbated by the power of AI inference and which was scarcely addressed by today’s decision by the Privacy Commissioner involving OpenAI.
Third, an enforceable duty to act responsibly that is tailored to the chatbot context. The architecture of chatbots, where output is generated in response to prompts rather than pushed through an algorithmic feed, makes age-tiered design feasible. A duty that mandates and audits developmentally appropriate design across different ages is a version of age-related regulation that fits the technology. A binary access cut-off borrowed from social media is not.
Now, I recognize that the political appeal of bans is obvious, but the case for them, especially in AI, is weak. We need to move on the harder and more useful work of building an effective Canadian model for AI regulation. I look forward to your questions.
The Chair: Thank you, Dr. Geist.
We will now proceed to questions from committee members. For this panel, senators will have four minutes for your question, and that includes the answer. Please indicate if your question is directed to a particular witness or all witnesses.
Senator Burey: I don’t know where to start. Hello. Okay. I will get one question out, at least.
I’ll speak with Mr. Proulx-Giraldeau. You talked about public trust, which you wanted to focus on, and you talked about this digital safety commission. I would like you to expand on that and also comment on the effects on democracy, “. . . if we don’t get these foundations right.” I am quoting you.
Mr. Proulx-Giraldeau: Thank you for the question. We stand at a very critical time right now. Earlier today, during the session, another witness quoted the KPMG study in 2025 on AI literacy and trust that people have in AI systems. In Canada, the portrait is not looking so good. When it comes to literacy and knowledge of AI systems, we’re forty-fourth out of forty-seven countries globally.
When it comes to trust in AI systems, nearly half of Canadians believe that the risks of AI outweigh the benefits. You compound that with kind of a trust crisis when it comes to federal institutions and democratic institutions in Canada.
According to Statistics Canada, about two thirds of all Canadians report low trust in the federal government. If we combine all of those effects together, I think it’s very clear that the state of things right now is insufficient to build trust with Canadians.
There have been several examples in the past of how real harms have come to pass to some people. This is why we, at Evidence for Democracy, are advocating for a digital safety commission as part of a renewed online harms act to really delve into these issues with a little bit more depth.
Senator Burey: Thank you. Dr. Geist, could you comment on this?
Mr. Geist: The online harms act was a tale of two bills. There was the element that many people thought provided a good starting point, which was premised on a duty to act responsibly that would then be fleshed out specifically for social media and targeted, exclusively, activities that are already criminal in nature.
It also included elements of changes to the Criminal Code and the Human Rights Act, which raised a lot of concerns around freedom of expression. It’s one of the things that bogged down the legislation itself.
I certainly think that it is important to see that legislative initiative renewed, especially focused, as I mentioned, on that duty to act responsibly.
Now, I would say that if we are thinking of trying to expand this beyond just the activities that were identified in the former Bill C-63 — let’s say, going to AI and thinking about how can we deal with it there — I would reiterate that AI and social media are different. We need to ensure that there is a granularity in terms of how we try to flesh out what those regulations look like.
In other words, a duty to act responsibly for Meta on the social media side is not the same as what it might look like for Anthropic or OpenAI with respect to its chatbots. I do have some concern that an attempt to have a one-size-fits-all to simply say, “Let’s come up with an online harms act and throw these standards at everyone,” runs some significant risks of having poorly fitting regulations.
Senator Burey: Thank you.
Senator Hay: Thank you all for being here. This is definitely a complex and complicated discussion.
Dr. Geist, I want to dive into the case for bans being weak. I was in Australia on some digital e-mental health work when that happened, and it was a real mixed reaction from people in the mental health world. The reality is that kids, young people of all ages, are everywhere in their digital world, including social media and AI, and 70% of young people are willing to give up their data for something in the digital world.
How do we use your idea, which I like, the transparency act, which is very similar to the digital commission that you were speaking about — how do we actually hold these multi-trillion-dollar market-cap companies accountable when what they are doing is building for their stakeholders, their shareholders?
If we do hold them accountable, how does Canada also put the policy in place to battle the potential pushback from the U.S. itself and them not wanting us to take on their hyperscalers?
Mr. Geist: Thanks for the question, senator. I’ll respond with a couple of things.
First off, just to emphasize with respect to Australia, when I say that the experience to date is that it doesn’t work, that’s coming from the eSafety Commissioner herself, who has identified that 70% of people who are on social media before are still online. There’s another survey that says the number is between 60% to 70%. But if you only get 30% in my class, you don’t pass.
The idea that it is effective — it’s simply not. In fact, there were many parents that said: “Where do I fit into all of this? I actually do want my kids involved.” We can get into some of that if you like.
Your question was, fundamentally, about how we can establish regulations that are effective involving some of these companies.
I would mention two things. First, there is a tendency — and I’ve heard it when looking at the transcripts around this committee and, frankly, just about all the committees that look at AI — to treat all these companies as somehow the same. They are all just big tech. They are not.
The positions, for example, that OpenAI has taken are very different from the positions that Anthropic has taken. Indeed, Anthropic started life, in part because it wanted to see more regulation, and it’s going around saying, “We are comfortable with a more regulated environment.” I think we need to be very cautious about seeing all these companies in the same light. They have different business models. You can see it taking place right now with respect to, say, Google’s Gemini versus Anthropic versus Meta versus OpenAI. They each have tried, or are trying, different things, and so thinking they are all the same, I think, is a mistake.
It’s a mistake to think that we can’t apply some of our laws here. I would point, frankly, to the decision or the fine literally handed out about six hours ago from the federal Privacy Commissioner and several other commissioners involving OpenAI and ChatGPT.
What you will find when you go through that decision is that the federal commissioner found, in a number of instances, well‑founded findings, instances where it said OpenAI was not compliant with Canadian privacy law. It found that OpenAI had addressed all of them.
Those that were well founded were addressed. In fact, the decision notes that OpenAI disagreed with some of the findings of the commissioner in terms of their analysis of what they had done as against Canadian privacy law, and they still made the changes.
The Chair: Thank you, Dr. Geist.
Senator Arnold: I’m with Senator Burey. It’s really hard to know where to start here. Thank you, Dr. Ponti, for your testimony. I don’t think I ever thought about those toys that kids use and the fact that they are unregulated. That has not come up in any of the testimony so far. Thank you for that.
I would like to focus on democracy today. Today, in The Globe and Mail, there was an article about foreign interference in the Alberta question right now, and I quote from the article:
Between late December and late April, references to Alberta separatism and various related themes, including talk of the province becoming a U.S. state and Canada failing as a country, rose sharply from known Russian content farms. During those four months, Alberta was the focus of 67 items produced and distributed by Pravda Network, nearly five times more than any other Canada-related topics.
So with this going on — and this is just one example — how do we get ahead of this? What can be done to undo some of this? Or is it too late and we just have to focus on going forward?
Mr. Proulx-Giraldeau: Thank you for the question. It’s an excellent question and, no, I don’t think it’s too late. Maybe it’s a little late, but better now than never.
It’s a broad question that you ask. The spread of misinformation online: how do we regulate this space? How do we, at the same time, protect free speech? How do we do this correctly? This is a difficult question. Many countries around the world are struggling with this, like we are here in Canada.
We have to think about regulating tech companies. Perhaps Professor Geist will have something to say about this issue.
Indeed, instead of putting the blame or the responsibility on users themselves, we have to think about regulating and controlling what these companies put out in terms of the algorithms spreading misinformation.
These tools and algorithms are designed with engagement in mind, not with safety or democratic principles necessarily. So we do have a responsibility to act in this space. Whether it’s labelling AI-generated content, I think that would be a wonderful first step, better than what we currently have, which is nothing; mandating algorithmic and content moderation is another way of going about this.
This is why I would go back to my opening remarks about this digital safety commission. A renewed Online Harms Act is more than due. This digital safety commission I was speaking about earlier would be basically the boots on the ground responsible for ensuring compliance with all of these companies.
Senator Arnold: Dr. Geist, do you have a response?
Mr. Geist: Sure. I have lots, but I probably don’t have that much time. I’ll make a couple of quick points.
First, misinformation and disinformation are not the same thing. We have this tendency to use this. We are concerned with misinformation and disinformation. There are lots of people who are misinformed and well meaning. They say things that are inaccurate, and they may get amplified. We have freedom of expression. There is no need or should be no expectation that we are going to find a law to shut that down.
Disinformation is something different. Disinformation brings with it some intent to misinform, often with some desire to have some kind of harm associated with it. There, we do need to ensure that we have effective rules to try to deal with disinformation and that the platforms respond effectively where they are certainly made aware of it and, in some instances, try to play a more active role even in terms of trying to deal with it.
Very quickly, this notion that we ought to be labelling AI content is true, but in some respects, it’s already happening. I can tell you, for example, I blog regularly and put on Substack regularly.
One of the images that I wanted to use was a nice image around surveillance, but it had some language in there that I didn’t really like. I put it into ChatGPT. I asked it to remove the language so it kept the image, created a slightly new image or removed that text. When I posted it to X, it labelled it as an AI‑generated image.
I think, in fact, there are incentives already in place for many of these services to engage in that labelling because users themselves want to know what information they can trust. We’ve heard that many people don’t have trust in the systems. There are already incentives to ensure that that takes place.
Could we do more? Of course. But I don’t think the notion that somehow these companies are just happy with AI slop and that as much of that as possible accords with their own business interests.
Mr. Reyes: Very quickly, this is exactly the kind of crisis I was talking about.
You can have on election eve a synthetic video appear that affects the result of the election, but fact checking takes days, and the voting is tomorrow. So what do you do? Many of the mechanisms were already expressed here.
But then there are many things that can be done, such as pre‑bunking, extreme transparency and truth hotlines during an election period.
There are mechanisms that are not necessarily AI-specific but that will address issues such as misinformation and disinformation related to elections. But the crisis is possible. I mean, it’s happening, and we have to deal with that.
Senator Senior: In the Canadian context, we have a neighbour to the south that is not a sleeping elephant. They are loud and boisterous. We also know that the border is an issue.
Based on everything you’ve said, including issues around consent — whether that’s possible; issues around transparency — whether or not that’s possible; issues around sovereignty — whether that’s possible, and as it impacts democracy. Do you feel confident that we can get this right?
I’ll start with you. Then I’d like to hear from Dr. Ponti as well as it concerns children.
Mr. Proulx-Giraldeau: May I ask what you mean when you say if we can get this right?
Senator Senior: In terms of what the Canadian government will do around strategy, legislation and regulation. Do you think we can get this right?
Mr. Proulx-Giraldeau: Doing something about it is definitely better than not doing anything. As we’ve heard before, inaction is a choice.
Senator Senior: It’s not about whether we should. Are you confident that we can, considering the fast pace that we’re at?
Mr. Proulx-Giraldeau: I think we can. You are right. There are things that are evolving fast and are outside of our control.
At the same time, there are many things we still have control over. We have heard from another witness this mention of pre‑bunking. Emphasis on literacy, education and critical thinking skills: those are all ways that we can actually — without necessarily involving those big tech companies, and regulating them, involving all the geopolitics, perhaps, behind this — protect the Canadian population against harms that these technologies may bring. That may be one element.
Senator Senior: Thank you. Mr. Reyes, you’re nodding.
Mr. Reyes: I fully agree. Geopolitics is affecting not just AI issues, but it will affect it.
Almost anything that Canada now is doing in terms of trade and investment policy, interactions with other countries, the middle powers, the whole variable geometry, are all triggered, and what’s impinging on it is the attitude of the United States. So whether it’s AI or other, that is almost an existential question. Maybe that is too much. But in terms of what can Canada do, and how much agency Canada can exercise in this particular domain, it applies to all domains of policy at the moment. That is the challenge for Canada.
Can it succeed? I think it can succeed with the right leadership, finesse and the right cooperative frameworks with other countries that are confronted with the same agency issue.
Dr. Ponti: Thank you for the question, Senator Senior. I can certainly speak on behalf of the CPS and the Child Health Policy Accelerator.
To answer your question: absolutely, we can get it right. Are we going to get it right tomorrow? Maybe not. But we don’t want perfection to be the enemy of the good enough. Absolutely, we are moving in the right direction.
These concepts of age appropriate by design and being flexible with emerging technologies all have to be built into this legislation. We’re talking about having an independent, online safety regulator. That could provide oversight, so let’s get something started and we can have further iterations as technologies advance and as our understanding advances.
As for the restriction issue, I agree, there are so many definitions and differences amongst different technologies. I certainly feel that we have to start somewhere. Certainly, restriction for the youngest of our population is only going to benefit them.
For example, AI toys: They are not beneficial to children’s development at all. I think it makes sense that we take a hard line. We regulate toys, stuffed animals and pyjamas; we should be able to regulate AI toys.
Senator Senior: Thank you.
Senator Petitclerc: Thank you. I have a question for Dr. Ponti as well, and maybe if I have time for you, Mr. Geist. You talked about growing concerns and you talked about youth and growing judgment, impulse control, long-term cognitive development, mental health and relationships.
We hear in the population often enough, and we hear in this chamber and in the other place that it is the parents’ responsibility to manage screen time and access to AI; it is your job as a parent to take care of that. Is it fair to put that responsibility on the parents?
Dr. Ponti: Thank you, senator, for the question. It is absolutely unfair to put it on any one parent, one family, one youth or one child. This is so much bigger than anyone.
It is because of the design of these products. They are designed to hook us with addictive designs. Because it is so unregulated, what parent is there 100% of the time hovering over their child while they are navigating the online world? It is unfeasible.
Also, even if you put screen time limits on one child, they are growing up in a family. They are growing up with friends and siblings and cousins, and they are exposed to it. Even if they are not directly using the technology, they are exposed to it. So I do not think it is fair for parents to have all of the responsibilities.
Absolutely, you should develop a plan for your family household, and we talk about that in our position statements to the Canadian Pediatric Society. But this is a much bigger issue.
Senator Petitclerc: Thank you.
If I hear you correctly, you favour regulations over a ban. My question then is this: How far or how strict should those regulations be, keeping in mind that my angle is for youth and kids, safety and knowing that everything goes so fast? There is research being done, but it is so new that we don’t know the real impact yet.
When it comes to regulation, should we go with a principle of caution to be on the safer side of things and reopen if needed when we have all the research and data coming back in a couple of years? Where do you see it?
Mr. Geist: I’m glad you raised the issue of the evidence in the early days because the reality is that the National Academies of Sciences, in their major report on social media a year and a half ago, found that, frankly, the notion that there is strong causation is not there. There may be a correlation in terms of harms, but what they largely found, even on the social media side, is that you have roughly, let’s say, 10% of users for which there are clear harms; 10% of users where there are clear benefits, where they find communities that they don’t otherwise have. And frankly, the majority of users for whom there may be some small, negative correlations in some instances, some small, positive correlations, but there were other factors that were far more pronounced. That is where the evidence has been going on social media.
On AI, it is way too early to suggest that, given how fast this has moved and how fast these AI chatbots are developing, we have strong evidence to suggest very clear-cut ultimate impacts. We can be concerned about it. Screen time is a good example. We have longstanding data that talks about concerns with respect to screen time. That’s not an AI chatbot issue, per se; that’s a specific issue around the use of these technologies more broadly.
The Chair: Thank you, Dr. Geist.
Senator Osler: Professor Geist, in your March 2026 appearance at the Standing Committee on Industry, Science and Technology, you cautioned against the risk of rapid and unbalanced AI regulatory frameworks and recommended that the government should prioritize passing modernized privacy and data governance laws.
For this committee’s benefit, could you please elaborate, first, on the specifics of what you would like to see in modernized privacy and data governance laws? Second, what would a regulatory framework that balances innovation and safeguards look like?
Mr. Geist: That is a great question. I’ll try and do justice to it.
On the privacy side, there are at least two elements we should be thinking about. One is that I don’t think anybody at this stage disputes or takes issue with the notion that Canada’s privacy law is out of date. We’re now dealing with a privacy law that is more than 25 years old.
We have seen in other jurisdictions, notably out of the EU, standards that are included that are simply not found in Canada. We could do worse than just modernizing it and keeping with what we see in other jurisdictions, whether that is around issues like the right to be forgotten, which certainly involve order-making power and penalties, which are virtually entirely absent from the federal system as it currently stands.
In fact, the decision today from OpenAI highlights what is an emerging fragmentation when it comes to privacy because the federal commissioner came up with a set of findings. We had B.C., Alberta and Quebec come up with, in some instances, different findings, so we have a problem with just basic modernization.
We also have — and I did a piece in The Globe and Mail about this — a real concern with respect to this notion of the ability for AI to re-identify and de-identify data. It is worth expanding on what that means.
In many instances, what organizations have done is said, “We’re collecting the data and we’re going to strip out any of the identifiers from there and so we can’t trace the information back to the individual.” That is one way to try and comply with these laws.
The problem is that, with AI and inference, it is not the large language models; it is now the ability for AI to go out and access different pools of information that are located in all kinds of places.
Its abilities to take these disparate pieces of information and, in a sense, put Humpty Dumpty back together again, take these pieces of information that on their own were not identifiable and were not designed to be identifiable, but it can use the inference and the power of the technology and can begin to identify what was previously not identifiable information. We need privacy laws that begin to think about that because they have been very reliant on that. There is a lot we could be doing on the privacy side.
On the AI side, in terms of what some of the data governance-related material is or what this looks like, as I said off the top, we could do worse than ensure that there is effective transparency because too much of this debate and discussion take place with opacity from the companies themselves. We don’t know enough about the algorithms or policies. Mandating greater transparency should be basic table stakes, and I do hear increasing support for this notion of some sort of duty of care or a duty to act responsibly.
I’ll acknowledge that not everyone is going to agree on what that means, but it provides an effective starting point to have the discussion about what it means to establish some kind of regulatory framework that serves the needs of Canadians and is also cognizant of some of the challenges we face to remain competitive in the environment.
On the privacy side, Japan has lessened their privacy protections. The EU has lessened their privacy protections. The U.K. has lessened their privacy protections. They are all moving in a direction that says they want to be leaders when it comes to AI. So when senators ask, “Are you optimistic about this?” I’m not that optimistic, not because I don’t think we’re smart, but because these require very hard policy choices.
It is not enough to say, “I favour privacy” or “I favour Canadian content in the data.” You have to make really hard calls, and I’m not entirely convinced that our current system is all that receptive to making those calls.
Senator Muggli: Thank you for being here. Mr. Reyes said earlier that AI can act as a crisis accelerant. That resonated with me, especially since I’m involved in a forest fire study right now. Throughout history, societies have failed to appreciate or react to the downside risks of new technologies unless something serious happens. We clearly need appropriate regulation, but I want to think a bit bigger and think about crises.
What would an effective AI emergency preparedness framework look like in Canada? I ask this because I’m thinking about AI Chernobyl. It could happen. It would be a major failure that exposes how unprepared our institutions are. Is this a concern of yours? If so, how do we prepare for that? I’ll go first to Mr. Reyes.
Mr. Reyes: From the very start, we have to do an audit of the crisis preparedness framework we already have. A crisis can happen today. If you do an audit of what we already have and how we respond to humanitarian crises, natural disasters, et cetera, and then understand if these frameworks can be adapted for AI.
It probably cannot in the current state because of the speed at which an AI crisis can unfold, but we have to think about it from that point, and then build up on what we can do. Biosecurity, for example, is a well-ignored area where there could be a crisis. So, a malicious actor or reckless lab uses AI to optimize pathogen design, evade screening or generate dangerous wet lab instructions. We can have a lot of imaginative scenarios. So we need to then establish a kind of AI biosecurity verification network and put up frameworks that would essentially prevent that. Mandatory screening for DNA synthesis orders and things like that can be implemented with the idea that an AI 9/11 can potentially be prevented because an AI 9/11 will happen soon enough.
Mr. Geist: I actually think you can make the case that, over the last year or so, there is the prospect that we face a crisis with respect to digital sovereignty or data sovereignty, especially when we think of the challenges with our neighbour to the south.
We have envisioned our response to that crisis almost entirely through the lens of “Let’s create Canadian-based infrastructure and that is will be the mechanism to address the issue.” The reality is that it doesn’t matter, in terms of your sovereignty, if your server is in Gatineau, so long as your laws are not effective to ensure that they apply. The easy part, so to speak, is to cut the cheques and say that we’re building up sovereign based infrastructure. The harder part is establishing the kinds of frameworks around sovereignty that will provide the guardrails or safeguards that we need to deal with what many people see right now as a crisis.
The Chair: Thank you very much. This brings us to the end of this meeting. I would really like to thank all the witnesses today for your time and for being with us today. This has been a fantastic two panels, and we have enjoyed it. We have learned a lot.
I remind members to review the draft outline summary document distributed this morning ahead of tomorrow’s meeting so that you come ready to give draft introductions on our report. Thank you all.
(The committee adjourned.)