Science and Research Committee on March 23rd, 2026

[Wyatt Tessari L'Allié]

That's a very good question.

I'll provide some context. Most cutting-edge research on artificial intelligence is done in Silicon Valley, often by Canadians, but OpenAI or Anthropic pays them millions of dollars. So Canadian AI research isn't really on the cutting edge anymore. There are some exceptions, but for the most part, that's where it's happening.

However, Canada is still in a very good position to advance research on AI safety. I know, for example, that initiatives like Mr. Bengio's and the LoiZéro organization, and all the work at the Canadian Artificial Intelligence Safety Institute, are moving in the right direction.

The problem with scientific research on artificial intelligence is that AI is advancing very quickly. Research projects often take years. We could have artificial intelligence that is smarter than humans in 18 months. So, if a research project begins today to look at an ethical aspect of AI, for example, or an aspect of controlling AI agent systems, by the time it yields results and is regulated, it will be too late. That's why we're increasingly in a situation where the most responsible thing to do is slow down or ban AI agents systems that are too dangerous, at least until safety research catches up.

[Mario Beaulieu, Bloc, La Pointe-de-l'Île, QC]

Could an independent body help decide which system should be banned?

[Wyatt Tessari L'Allié]

Yes. For example, the Canadian Artificial Intelligence Safety Institute is a step in the right direction, but it has a very small budget and it doesn't have the mandate to monitor autonomous agent systems. So a reorganization would be in order and the mandate would have to be changed. In addition, the institute should be given more power so that it can have an influence, and not just make information public without being able to react to it.

So I would say that this institute should be strengthened, and we should work with institutes around the world. The one in the U.K., the AI Security Institute, is probably the best example. They do tests on leading-edge models, and I would say they're among the most aware of what's going on.

[Mario Beaulieu, Bloc, La Pointe-de-l'Île, QC]

Thank you.

Dr. Michaud, how do you ensure that francophone teams have equitable access to platforms, tests and assessments? I'm thinking in particular of French-language services or committee membership.

[Stéphanie Michaud]

At BioCanRx, we subsidize all the universities in Quebec. In addition, we subsidize about a third of the clinical trials that take place in Quebec. In terms of assessing the different projects, we have international committees. Committee members come from all over the world, because we seek out certain types of expertise. Committee meetings do take place in English, but all the universities in Quebec still participate. All those who have medical schools, such as the Université de Sherbrooke, Université Laval, Université de Montréal and McGill University, participate.

BioCanRx currently subsidizes about nine research teams in different types of translational programs.

[Mario Beaulieu, Bloc, La Pointe-de-l'Île, QC]

They still have access to platforms and all the tools in French, don't they?

[Stéphanie Michaud]

Yes. Our website and our communications are always in both languages.

[Mario Beaulieu, Bloc, La Pointe-de-l'Île, QC]

That's great.

On the whole issue of microdata collected in secure environments, what clinical and performance data from immunotherapy projects could be shared anonymously to allow for independent evaluation of results and biases?

[Stéphanie Michaud]

If you want to talk about the clinical trial results, of course they're available. The data is collected in the various hospitals where the clinical trials take place. Those kinds of results don't give us access to patient names. We only have access to the overall clinical trial results that we fund.

So everything is well protected through the various systems in the hospitals we work with.

[Mario Beaulieu, Bloc, La Pointe-de-l'Île, QC]

That's excellent.

What kind of governance adaptation between organizations like the Canadian Institutes of Health Research or the Social Sciences and Humanities Research Council would reduce fragmentation and improve the traceability of research funding decisions?

[Stéphanie Michaud]

That's a great question. Of course, that's why we're here today.

We really need to look at a new governance system to properly monitor development. In the case of BioCanRx, it's the development of a product from discovery to the clinical trial. Right now, there's no way to have a new system, but non-profit organizations like ours, which are funded by the strategic science fund, are used to manage the immunotherapy development pathway.

Better governance between organizations is really something that should be considered for developments related to the various diseases patients are grappling with across Canada.

[Mario Beaulieu, Bloc, La Pointe-de-l'Île, QC]

Thank you.

My next question is for you, Mr. Annan—

[The Chair, Liberal, Salma Zahid]

I'm sorry for interrupting, but your time is up. Thank you.

This ends our first round of questioning. We will start our second round of questioning with MP Holman for five minutes.

Please go ahead.

[Kurt Holman, Conservative, London—Fanshawe, ON]

Thank you, Madam Chair.

This is a question for Mr. Tessari L'Allié.

The federal government needs stronger AI governance. There have been repeated concerns about delays, limited consultation and a lack of clear legislative direction.

Your organization made a submission to the national AI task force. Do you feel your key concerns were meaningfully reflected in the government summary? If not, what was missing?

[Wyatt Tessari L'Allié]

I would say that, broadly, no, our recommendations were not reflected, partly because there were 11,000 submissions and ours was only one of them.

The biggest piece was our message to government for the last three years, which has been this: Think ahead. AI is moving very quickly. If it takes government a few years to put something in place, you need to think about where AI is going to be in a few years. That's why you have to follow what every major AI company is saying. They are saying, “We're building 'smarter than human' AI systems—systems that can do everything a human brain can do, only faster, better and cheaper.” We need to be ready for that.

On the legislation front, the AI and data act was probably the last chance to get something in place before “smarter than human” AI. You can still pass a bill like Bill C-5, which was rammed through, but that's the speed at which you would have to move. You would also have to make it very flexible so the regulator can regulate very quickly, because if there's a new capability coming out every week, you're going to need a new regulation every week. That's the speed at which you need to move.

[Kurt Holman, Conservative, London—Fanshawe, ON]

Thank you, sir.

You raise concerns about the lack of oversight for rapidly evolving AI systems, including open-source models. Is the federal government behind on developing its AI policy and safeguards for Canadians?

Then I have a follow-up question. Have you seen any progress with the federal government AI policy ever since your appearance and warnings at the industry committee two years ago?

[Wyatt Tessari L'Allié]

I will say there has been increasing awareness over time, and every country is far behind in this, so it's not just Canada.

I'm sorry. I forget the other question.

[Kurt Holman, Conservative, London—Fanshawe, ON]

Is the federal government behind on developing its AI policy and safeguards for Canadians?

[Wyatt Tessari L'Allié]

Yes. Everyone's behind. Even the AI companies at the front end, at the frontier, are saying that they're going too fast to be able to often test these things sufficiently before they roll them out. There's a huge competitive pressure right now to go as fast as possible, because the company that doesn't shift, you know, is the company that falls behind. That is increasingly dangerous because we're seeing, for example, Anthropic, which is considered one of the most responsible organizations in the field.... They recently publicly reneged on their pledge to not build unsafe systems, and they did so because they realized that if other people are building unsafe systems and they're not, they'll fall behind.

That's where you really need governments to be able to see this happening and step in now. At this point, you don't have five years to put a law in place. You're going to have to slam the brakes on the stuff very quickly if you want to stop the worst outcomes.

[Kurt Holman, Conservative, London—Fanshawe, ON]

Would you say in a quick overview that Canada is not fully prepared for AI, that the government has not been transparent or inclusive in how it's developing policy and that, at the same time, our funding system is not delivering results for Canadians—i.e., the government is behind, experts are not fully heard and there are real, unaddressed risks? Would you agree or expand on that?

[Wyatt Tessari L'Allié]

That's roughly correct in the context of every other country being in the same boat. As mentioned, time really is running out. The experts we talked to in the frontier labs are saying that this thing's happening much faster than people think it is. The latest breakthroughs we saw in January and February, where we had AI agents.... Literally, companies are buying extra computers for their employees so that their employees can have multiple computers. Each of those computers is running an AI agent on its own, kind of like its own AI employee.

That is what's happening. It's going to transform research as well. I haven't really gotten into this yet, but a lot of the fundamental assumptions we have about how the economy works, how human labour works, how human cognition works, and the value of human knowledge are about to get very rapidly disrupted. That's why we need time to work through that. There's a lot of opportunity there to work through if we do get through it, but it's going to require some very hands-on governing in the very short term.

[Kurt Holman, Conservative, London—Fanshawe, ON]

Do you feel that Canada should require mandatory safety testing or certification before high-risk AI systems are deployed?

[Wyatt Tessari L'Allié]

If Ford and GM have to do testing before they can put a car on the road, the same thing should happen with an AI system.

In the 2010s, you had these single-purpose, machine-learning tools. You were training your model to recognize images or to work on a bigger science project. Yes, you need some regulation, but it's not too big a problem. Chatbots are a similar problem, and you're starting to see issues with creative industries.

However, we're now in a very different era. We don't just have AI that can talk; we have AI that can act, AI that is buying and selling stuff, AI that is making phone calls, AI that is interacting with people, as a human being would. That is a whole other level of risk, and that's where we have to deal with it very quickly because we're waiting for an accident to happen, basically.

[Kurt Holman, Conservative, London—Fanshawe, ON]

Thank you.

[The Chair, Liberal, Salma Zahid]

Now we will proceed to MP McKelvie for five minutes.

Please go ahead.