Canada Wants Sovereign AI. But Sovereign for Whom?

Canada is suddenly very interested in sovereignty. The word appears everywhere in conversations about artificial intelligence: sovereign AI, sovereign compute, sovereign cloud infrastructure and sovereign data.

Canada’s new national artificial intelligence strategy, AI for All, identifies sovereignty as a central priority. It calls for Canadian capacity in compute, data, talent and infrastructure as the country seeks greater control over the technological systems on which it increasingly depends.

These are important conversations. But the questions underneath them are not new.

An Old Question in a New Setting

Who owns information? Who controls it? Who determines where it is stored, how it can be used, who can access it and what happens when its original purpose changes?

First Nations have been asking versions of these questions for decades. The First Nations Principles of OCAP®, Ownership, Control, Access and Possession, emerged from a history in which information about First Nations people and communities was too often collected, interpreted, stored and circulated by institutions outside those communities, without First Nations exercising meaningful authority over what happened to it afterward.

OCAP® changed the question. Instead of asking only whether information was collected ethically, it asks who holds authority over it. That distinction matters enormously in the age of artificial intelligence.

A First Nations framework, Not a Universal One

It is also important to be precise. OCAP® is specifically a First Nations framework. It does not stand in for the distinct laws, governance traditions and approaches to data sovereignty of Inuit or Métis, nor does it erase the distinct laws, protocols and governance traditions of individual First Nations. Even among First Nations, its expression is not identical everywhere. The First Nations Information Governance Centre emphasizes that Nations assert these principles according to their own worldviews, traditional knowledge and protocols.

That distinction is not a limitation of the framework. It is part of the point.

Sovereignty does not mean applying one model everywhere. It means having the authority to determine the model in the first place.

Consent to Collect Is Not Consent to Train

Much of the public conversation about responsible AI centres on familiar ideas: privacy, cybersecurity, transparency, bias and consent. All matter. But artificial intelligence complicates what consent means over time.

Consider information contributed to a community health study. Recordings made as part of a language revitalization initiative. Oral histories deposited in an archive. Stories shared through university research. Information provided to a government so that citizens can receive programs or services.

Permission for those purposes does not automatically constitute permission to use the same information to train an AI system.

Digital information can appear infinitely reusable. Governance reminds us that it is not infinitely available.

Once information has been digitized, copying it is easy. Combining datasets is technically possible at enormous scale. Information collected for one purpose can become attractive for another purpose that may not even have existed when it was originally gathered.

AI intensifies that problem. Information may no longer simply reside in a database. Once used in AI development, it can shape systems that generate language, images, classifications, predictions and representations long after the information was originally collected. The consequences of secondary use can therefore become harder to see and potentially harder to reverse.

When the technology changes, does the original consent travel with the data?

What the Four Principles Ask

OCAP® gives us a useful way to understand why the answer cannot simply be assumed.

• Ownership considers the relationship between First Nations and their collective cultural knowledge, data and information.
• Control concerns First Nations’ authority over research and information management processes that affect them, from planning and collection through management and use.
• Access means First Nations must be able to access information and data about themselves and their communities regardless of where it is held. It also includes the right of First Nations communities and organizations to manage and make decisions about access to their collective information.
• Possession makes the conversation tangible. While ownership describes the relationship between a people and their information, possession concerns physical control of the data, a mechanism through which ownership can be asserted and protected.

That final principle may become increasingly important in the AI era.

Where Does sovereignty Live?

Canada’s emerging approach to AI recognizes that infrastructure matters. The federal government’s national AI strategy says sovereignty in the AI era depends upon compute, cloud and connectivity, the data that fuels AI systems and the people who build and govern them. It also acknowledges Canada’s current dependence on foreign cloud platforms and infrastructure.

There is an important parallel here. If Canada considers control over technological infrastructure relevant to its own digital sovereignty, what does control over infrastructure mean when we talk about First Nations data sovereignty?

The parallel is not an equivalence. Canadian digital sovereignty and First Nations data sovereignty arise from different political, legal and historical relationships. First Nations sovereignty cannot be reduced to a variation of Canada’s concern about foreign-owned cloud infrastructure.

But both expose a limitation in treating data governance as merely a question of privacy or security. Control matters.

What meaningful control looks like will depend on the Nation. For one Nation, possession might mean servers controlled by the community. For another, it could mean a regional data centre governed by Indigenous communities. Others may choose trusted partnerships, federated systems or commercial cloud services governed through carefully negotiated agreements.

The important question is not whether every Nation builds the same infrastructure. It is whether Nations have meaningful authority to determine what arrangement serves them. That includes the authority to determine that certain information should never enter an artificial intelligence system at all.

The technological instinct is often to ask how information can be made more accessible, interoperable and useful. Indigenous data governance introduces another legitimate question:

Should it be?

Not all knowledge becomes more valuable when it becomes more available. Not everything that can become training data should become training data.

Representation Is Not the Same as Authority

There is increasing recognition that Indigenous Peoples should be represented in conversations about artificial intelligence. Representation matters. So does addressing AI systems that reproduce stereotypes, misunderstand Indigenous histories, flatten distinctions between Nations or generate inaccurate representations of Indigenous languages, cultures and knowledge.

But representation cannot substitute for governance. A system could become exceptionally good at generating content about Indigenous Peoples while Indigenous Peoples exercise very little authority over the information that contributed to that capability.

In fact, the better these technologies become at generating or reproducing cultural material, language and knowledge, the more urgent the governance questions become.

Questions the Technology Cannot Answer for Itself

Who authorized that information to become training material?

Were the people who shared it told that this was a possible future use?

What happens when information originally shared in one relationship moves into another?

Can consent be withdrawn?

Can a Nation establish conditions that follow information beyond the institution that originally collected it?

Can some knowledge remain available for community purposes while being unavailable for commercial AI development?

Who benefits economically when culturally specific information contributes to valuable technologies?

And who carries the responsibility when those technologies get that knowledge wrong, not only for the consequences experienced today, but for what those decisions leave to generations yet to come?

For Haudenosaunee peoples, that responsibility carries a longer horizon. Thinking seven generations ahead asks us to look beyond immediate technological benefits toward the relationships to knowledge, authority and responsibility that our decisions create for those who will inherit them.

These are not merely technical questions. They are questions about authority, relationships and responsibility.

There is an important difference between asking how Indigenous Peoples can be better represented within artificial intelligence and asking what Indigenous governance requires of artificial intelligence. The first question usually begins with the technology. The second begins with the people whose information, knowledge and relationships may be affected by it.

That changes where the conversation starts.

Before the Model

Too much of the conversation about responsible AI begins after the architecture has already been built. A model exists. A dataset has been assembled. A platform has been deployed. Then we ask how to make it ethical.

How do we reduce bias? How do we make the system more representative? How do we consult affected communities? How do we create safeguards around what has already been built?

Those are worthwhile questions, but First Nations data governance asks us to move the point of governance earlier: before the model exists, before the data is scraped, before an archive is digitized, before a researcher uploads a dataset, before culturally significant material becomes machine-readable, before someone checks a box marked “I agree,” and before Indigenous knowledge becomes another resource powering the next technological revolution.

This is where OCAP® offers something important to contemporary debates about artificial intelligence. Not a technical specification. Not a universal Indigenous AI framework. And not a checklist to be added to an AI ethics assessment after development is already underway.

It offers a different starting place: the relationship between peoples and their information, and the authority that should accompany that relationship.

The First Nations Information Governance Centre describes possession as the mechanism through which ownership can be asserted and protected. That idea becomes especially consequential when the infrastructure holding data, and the systems learning from it, may be owned and operated far beyond the Nation itself.

AI therefore makes an old governance question newly urgent: who has the authority to decide what happens next?

Canada is investing heavily in the idea that sovereignty matters in artificial intelligence. OCAP® reminds us that sovereignty cannot be reduced to where servers are located or who owns the company operating them. It is also about relationships between people and information: who can say yes, who can set the conditions, who can change those conditions and, importantly, who retains the authority to say no.

As governments, universities and technology companies race to define responsible AI, First Nations data governance asks us to begin somewhere else: What if sovereignty begins before the technology does?

---

Editor’s Note: On the National Day for Truth and Reconciliation, read the Canadian International Council’s Statement on Reconciliation with First Nations, Métis and Inuit Peoples to learn more about CIC’s commitment to reconciliation.