Canada needs our own Mythos AI – and we can achieve that

Open this photo in gallery:

David A. Johnston is director of the master of supply chain management program at York University’s Schulich School of Business, where Hjalmar K. Turesson is director of the master of management in artificial intelligence program.

Last month, leading Silicon Valley tech firm Anthropic announced a new version of its proprietary large language model called Mythos.

Because of its potential to produce malicious code that makes mission-critical systems vulnerable to cyberattacks, it was initially only released to 50 organizations – all based or with operations in the United States and subject to U.S. law.

On Tuesday, Artificial Intelligence Minister Evan Solomon said that, through the Canadian Centre for Cybersecurity, Ottawa has become part of Anthropic’s preview campaign. This will allow some Canadian companies to access Mythos.

That is a welcome move, but it does not negate the fact that Mythos and other leading AI models are American technology. These firms arguably have a chokehold on the information technology used by most Western democracies.

Opinion: Canada is not likely to become a conventional AI power. But here’s how it can still win

For Canadians concerned about our country’s capability to secure our own supply chains of data and power to process information, we propose two guiding principles. First, Canadian public- and private-sector organizations should adopt and modify open-sourced AI models for domestic AI applications. Second, the physical infrastructure used to run and store these models should be located in Canada and owned by Canadians.

Currently, the commercial LLMs that underpin the AI used by most consumers, businesses and governments in Canada have primarily been developed by three large American corporations. The U.S. CLOUD Act leaves the content on their servers accessible to the U.S. government, irrespective of where the servers are located. This is unacceptable from a national-security perspective.

However, recent developments in AI mean training a new open-source LLM can be done cheaply and securely, leveraging the same publicly available digital sources used by their commercial counterparts.

Chinese startup DeepSeek released its landmark, low-cost, open-source model in January, 2025. Similarly, there has been an open-source model released about three to four months after every major commercial LLM release. These models closely match the capabilities of those released commercially.

An open-source model as capable as Mythos should be expected within half a year. In short, Canada can run and control its own models, which would be just as capable as the frontier versions, without investing the hundreds of millions to billions it took to develop models such as ChatGPT and Claude, or compromising our digital sovereignty.

Minister Solomon confirms Canada now part of Anthropic’s Mythos AI preview program

Digital sovereignty is about a country controlling its own data, the processing of that data and the resulting knowledge that it produces. It means securing our individual privacy and property, including the intellectual property we have developed.

Every Canadian has personal data that should be secured in Canada, away from the legal authority of other nations with potentially conflicting political values and commercial interests.

Canadians expect that their medical records are secured locally; that public education institutions run safe online spaces where the digital identities of students, faculty and staff are not mined for commercial exploitation or social abuse; and that transportation networks, pipelines, electrical grids and other infrastructure are secured against cyberthreats.

AI models will increasingly be used to address a range of challenges, such as managing weather-related supply disruptions, reducing internal trade barriers and increasing our capacity to import and export with a greater diversity of countries. These models and the cloud from which they emanate should be controlled by Canadians, subject to Canadian law and the political forces that shape it.

The good news is that organizations are taking steps toward independent control of AI models, the data they use and the resulting knowledge. Some universities have their own limited AI platforms based on open-source models.

While much has been written about the massive data centres built to power AI, newer models are becoming increasingly compact, requiring less hardware.

For example, an open-source model such as Google’s Gemma 4 E4B can be loaded and run on the average smartphone, yet it is more capable than the earlier version of OpenAI’s GPT-4, which required a data centre to run. The capability density of AI models, a metric measuring a model’s efficiency, doubles every three to four months. That also means less hardware and energy consumption.

Canada leads the G7 in its working-age population with college or university credentials. Both of us are educators running university programs whose graduates can help supply the manpower Canada needs to achieve digital sovereignty.

It’s important to keep those educated individuals in Canada once they graduate. They should be employed in new or existing Canadian organizations that can buy or develop digital services locally, rather than only facilitating the purchase of those products and services from international providers.

Canada can, and should, have its own Mythos and the infrastructure to run it.