During AICON Canada, I spoke with teams across the country about how they are actually adopting AI in practice. Historically, Canada has often felt a few years behind the UK and Europe in terms of adoption, but there are strong pockets of innovation emerging, particularly in Alberta and Nova Scotia, and a growing sense that momentum is building.
Fast forward a few days later, and Canada's national AI strategy, AI for All, has just been released. This strategy suggests that gap is starting to close.
There are three elements that stand out from a practitioner perspective.
- First, this is an adoption strategy. The focus is on scaling use across the economy, supported by investment in skills and partnerships.
- Second, there is a clear commitment to building a sovereign AI foundation, including domestic compute and reducing reliance on foreign infrastructure and talent.
- Third, adoption is directly linked to trust, with an emphasis on safety and public confidence.
None of this is entirely new. What is different is that it is being treated as a connected problem.
Trust as a delivery mechanism, sovereignty as control
One of the more important shifts in the strategy is how it treats trust.
Trust is not referenced as a simple principle or a value statement. It is positioned as something that enables or blocks adoption. Canada is quite explicit that without public and organisational confidence, AI will not scale.
This aligns with what we are seeing in delivery. Most organisations can build AI. What slows them down is the lack of clarity around how those systems are governed, what decisions are acceptable, and who is accountable when things go wrong.
Responsible AI sits directly in that gap.
Where it is treated as a late-stage exercise or a form of validation, adoption tends to stall. Where it is embedded into how systems are designed, decisions are made more quickly and with more confidence. The strategy does not solve this, but it does recognise that this is where the real constraint sits.
The second theme, sovereignty, is more structural.
The focus on domestic infrastructure and data control reflects a view of AI as something that needs to be operated and controlled locally, not just consumed. For organisations, this becomes a question of assurance.
If you do not control where systems run, how data is governed, or which legal frameworks apply, it becomes harder to make credible claims about safety, accountability, or fairness. Sovereignty, in that sense, is not separate from Responsible AI. It is part of how it is delivered.
How this compares to EU and UK approaches
This is where the Canadian approach starts to diverge.
The EU has taken a regulatory-first position. The AI Act sets out a comprehensive, risk-based framework with defined obligations depending on the category of system. That provides clarity and enforceability, but it can also introduce friction, fear of penalties and generally slow things down.
The UK has taken a more principles-led approach, relying on existing regulators and guidance rather than a single binding framework. That creates flexibility, but also leaves organisations to individually interpret what "good" looks like in practice.
Canada is taking a more operational route.
Rather than leading with regulation or guidance, it is focusing on the conditions required to make adoption work. That includes investment in infrastructure, a clearer link between trust and delivery, and an explicit focus on scaling use.
From a practitioner perspective, this reflects a shift in where the problem is understood to be.
It is not a lack of ideas, or even a lack of regulation. It is the difficulty of turning both into something that works at scale.
What this means for the AI market
The strategy does not introduce a new direction. It confirms one.
Adoption is becoming the central challenge. Trust is being treated as a condition for scaling. Sovereignty is increasingly tied to control and assurance.
These were consistent themes at AICON Canada, particularly when looking at where initiatives slow down.
Most teams are not blocked by capability. They are slowed down by the number of decisions that need to be made around governance, accountability, and acceptable outcomes. That only becomes more complex as systems become more autonomous.
What this strategy signals is that these are no longer internal organisational issues. They are now being recognised at a national level. For organisations, the implication is relatively straightforward. It will not be enough to demonstrate capability. There will be increasing pressure to demonstrate control.
And in practice, that comes back to a small set of things: how decisions are made, who is accountable for them, and how you evidence those decisions when put on the spot.