GC Data Conference 2026 - Day 1 - February 18, 2026
Disclaimer: The summaries and interpretations provided on this page are unofficial and have not been reviewed, endorsed, or approved by the Canada School of Public Service (CSPS).
Summary
Panelists:
- Nick Frost – Co-founder, Cohere
- Bailey Kacsmar – Fellow, Alberta Machine Intelligence Institute (Amii)
- Abdi Adid – Professor, University of Toronto Faculty of Law; Canada Research Chair in AI & Access to Justice
Why Data Matters
- Evidence-based policymaking is still not the default in government — it remains more of an aspiration than standard practice
- Data underpins innovation, sovereignty, public services, and economic competitiveness
Privacy & Public Trust
- Most data collected is ultimately data about people, not an abstract concept
- Big data allows synthesis of innocuous information into deeply personal insights (e.g., predicting pregnancy, health conditions) — existing privacy law doesn’t adequately address this
- Canadians rank among the highest in the OECD for distrust of AI and technology systems
- This distrust stems partly from Canada’s history (e.g., marginalized communities’ legitimate reasons to distrust government) and partly from a lack of clear recourse mechanisms when things go wrong
- Privacy is not about “hiding” — it’s about maintaining selfhood and control over how personal information flows in social contexts
- The “nothing to hide” argument is a fallacy; privacy is complex, cultural, and deeply human
- Europe (GDPR) has strong data protection; the US has almost none at the federal level; Canada sits in between with outdated legislation
Data Ownership vs. Agency
- “Ownership” is the wrong framework for data — unlike physical property, data is non-excludable (you can share it and still have it)
- Better framing: agency and dynamic control over how data flows
- Purely market-based “own your data” models risk creating a two-tiered system where only those who can afford it get privacy protection
AI Sovereignty
- Only 4 countries can build frontier AI foundational models: the US, China, Canada (Cohere), and France
- Only ~10 companies worldwide can build these models
- Sovereignty in this context = autonomy and agency — being in control of what happens to you at the individual, corporate, and national level
- Canada’s data sovereignty is both a matter of national security and an economic growth strategy (repatriating data value to support domestic industries)
- The US CLOUD Act allows US law enforcement to access data hosted in or routed through the US — a significant risk for Canadian data held with US cloud providers
- Canadian health data is legally protected from being hosted outside Canada — cited as a model for conscious, deliberate data policy
Canada’s AI Paradox
- Canada invented much of the foundational AI research (Geoffrey Hinton, deep learning, Toronto/Alberta/Montreal ecosystem)
- Yet Canada has the lowest rate of AI adoption among OECD/G7 countries (~12% at the firm level)
- Reasons include: legacy industries (telecom, extractive resources) with little incentive to change; low venture capital liquidity; insufficient labor mobility compared to the US; the gravitational pull of Silicon Valley
Cohere as a Sovereign AI Company
- Cohere deliberately chose to stay in Canada despite it being harder to raise capital and attract customers
- Now has a signed MOU with the Government of Canada to build and deploy AI within government
- Described as one of the few Canadian-founded tech companies to scale globally while remaining headquartered in Canada
- Nick Frost expressed discomfort with Silicon Valley’s model: addictive tech, youth anxiety, lack of social benefit — Canada should build its own version
Procurement Reform
- Government procurement was designed for hardware, not software or AI — it is fundamentally mismatched for iterative, continuously improving technology
- Public sector procurement cycles (2–3+ years) mean that by the time a technology is deployed, it is already outdated
- The MOU between the government and Cohere was announced publicly, yet no procurement vehicle existed the next day to actually use it — illustrating the disconnect
- Private sector analogy: you “date” slowly but once committed, the deal closes quickly; government does the opposite
- Recommendations: design procurement to anticipate updates, avoid over-specification of version numbers, hit the “turbo button” on procurement timelines
- Risk of a structural lock-out: public servants who don’t get early AI exposure may lose transferable skills; agencies risk falling permanently behind in service delivery capability
Closing Advice to Public Servants
Bailey Kacsmar
- Be purposeful — don’t adopt AI wholesale or scatter-gun it across all domains
- Understand your domain first; identify where AI genuinely adds value
- Misuse destroys trust, which is very hard to rebuild
Abdi Adid
- Be a credible objector — identify your genuine no-go zones (e.g., AI in criminal sentencing) and defend them rigorously
- But don’t let principled objections bleed into rejecting low-stakes, beneficial uses (e.g., summarization tools)
- Engage in the conversation about the future of work rather than opting out entirely
Nick Frost
- AI and programming are not beyond you — learning a little Python is more accessible than ever
- Treat learning to code like learning a language: hard but achievable and valuable
- Understanding how LLMs work at a basic level is within reach for everyone, even if building them is not
- Canada’s low adoption rate is a problem — more people need to actively explore where they do want to use AI