AI Transparency In Canada

On July 23, 2026 the Canadian federal government issued an open invitation to Canadians to “Have your say on advancing AI transparency in Canada”.

Knowing my interest in AI my daughter encouraged me to respond to the invitation. Over the past couple of years I’ve grown to appreciate that one of the greatest things about a democracy is having a say. So I decided to respond to the invitation.

The consultation questionnaire is extensive. It addresses six areas of AI:

1. AI Generated Content

2. AI Interaction

3. Information about AI Systems

4. AI Incidents

5. AI Agents

6. Broader Considerations

I’ve been finding this framework interesting when looked at through an education lens. Some areas I’ve been tracking and have thoughts on, others I’m aware of but have not explored deeply so don’t have informed feedback.

The feedback form has a lot of questions in each area. I downloaded the form they are using to invite feedback and slowly worked on a personal response to the questions from an open point of view. As I did so I kept thinking it would be awesome if a larger group response came from the Canadian open education community. I think a group response would be more thorough, convey a range of perspectives, and carry more weight. I think the government needs to start hearing from open educators and we need to proactively try and shape issues and policy relevant to our work. All that led me to wonder - Is there any interest in collaborating on a collective response?

I’m part of the OEGlobal Connect forum a space where discussions on all things related to open can take place so I posted an invite there. A couple of people expressed interest in responding. By the time I posted the invitation the due date for responses was fast approaching leaving little time for organizing the logistics of a collective response so while it was great to see a couple of people interested in the end I simply shared some of the questions and my responses in the forum.

The first section of the feedback form dealt with AI-Generated Content. The first question was:

Question : Would it help you trust what you see online if you could tell whether something was created by AI or by a person? When and for what kinds of content (e.g., image, video, audio) is it most important to know if something was created or modified by AI?

Pauls Answer: I am particularly focused on the education sector. In that context it is very important to know whether something was created by AI or a person, or some combination of both. This applies to all kinds of content.

Such a requirement is not that different from how traditional media is currently handled in education. Authorship is stated up front and verifiable. Those who play an intermediary role such as publishers are similarly noted upfront along with contact information, dates of publication, and licensing rights. Content created in an education context by all parties, authors, publishers, students and educators requires references. This practice is what creates trust and responsibility.

Currently AI does not follow any of these practices. AI technologies are not transparent around their source data. This seems largely to be based on the (illegal?) practices AI developers used to acquire data for training their AI models. Disclosing data invites scrutiny AI developers facing copyright litigation don’t want to expose themselves to. Amazingly this practice of non-disclosure has been widely accepted without regulation. Imagine if content generated by a publisher, student or educator did not divulge sources. It would be immediately flagged as unreliable and unacceptable. Imagine if food at a grocery store had no label! It seems odd that we have regulations around food safety requiring disclosure of ingredients but none around AI.

The absence of these practices immediately results in trust issues. AI technologies at this point are black boxes with little to no information disclosed on the underlying sources used to generate AI content and frequently no or fake or low quality referencing. Without addressing this fundamental requirement AI will automatically be untrustworthy.

The Data Provenance Initiative and articles like Sufficiently detailed? A proposal for implementing the AI Act’s training data transparency requirement for GPAI and the AI Transparency Template Blueprint offer a range of approaches.

But the issue is not simply about “content” it is also about the production process used to create the content. In food production there are regulations that specify what is required for food to be designated “organic” or “fair trade”. In addition, there are legally defined and protected geographical regions defining where only certain food products can be grown and associated laws that prescribe rules that must be followed around how the food is made within those boundaries. AI is currently completely devoid of such rules and regulation. This has resulted in AI slop and AI hallucinations.

In education it has also resulted in transference of cognitive function from human to machine with few or no guardrails and little consideration of the consequences. Cognitive offloading to AI is not a positive step forward for education or society overall.

Creative Commons notes “Attribution has always been a cornerstone of the commons. It supports participation, enables transparency, and allows knowledge to be traced, evaluated, and built upon.

Today’s AI ecosystem is eroding this norm. Most generative systems do not meaningfully acknowledge the sources they rely on. As AI increasingly mediates access to knowledge, this has serious consequences: loss of provenance, reduced trust, and fewer incentives to share.”

The second question in the AI Generated Content asked:

Question: What would best help Canadians determine when content is AI-generated (e.g., invisible or visible watermarks, disclaimers, provenance metadata)?

Pauls Answer: AI technologies have tried to push the notion that AI is human-like by creating personifications through things like using human names (e.g. Claude) or calling them “agents” (personification akin to being a travel agent). But AI is not human like. It is simply a statistical prediction technology.

A better analogy might be to consider AI like a country. We are used to seeing “Made in …” designations specifying country of origin. AI generated content could be simply designated “Made by AI”. This designation could contrast with the default assumption of “Human made”.

In the capitalist marketplace we are also used to seeing brand names associated with content – Nike, Apple, etc. So the designation could be “Made by (brand or application name) AI”. This might provide further clarification as to quality or trustworthiness. I think there are opportunities to express a broader range of provenance meta data too.

A significant challenge lies when content is produced by a mix of AI technologies, or content created by a mix of AI and humans. Clothing labels might provide some direction where fibre content of clothing is often designated using percentages (e.g. 50% wool, 50% cotton).

Attribution for AI provides a starting point for labelling.

From Prompt to Practice: A Framework for Transparent GenAI Use in Higher Education provides a framework for education that could also be relevant for use in other contexts.

It is also interesting to see certain businesses adopting to exclude AI and embrace human creation. The Noun Project has released new Terms of Use specifying “AI-generated content may not be submitted to Noun Project’s icon collection, which will remain human-created.”

AI slop batters book publishing illustrates what can potentially happen when content is not appropriately labelled as AI generated. It also reveals the complexities as content can, and often is, a mix of human and AI generated. Such issues raise significant questions about the desirability of AI and call into question the constant hype of how AI is a good thing – including hype in the vision set out in Canada’s National Artificial Intelligence Strategy: AI for All.

I think the public currently questions the benefits of AI. Benefits to the public are still more hypothetical than real.

Here are the third and fourth questions and my answers:

Question: Who in the AI value chain (developers, deployers, or others) should be responsible for providing transparency around AI-generated content? Why?

Pauls Answer: All players in the value chain have an obligation to provide transparency around AI-generated content. Going back to my earlier reflections on education every player in the education value-chain is expected to give attribution and acknowledge sources. They are also responsible for the legality, accuracy and validity of the content they produce. This should be true for AI too.

Focusing only on transparency is to limiting Developers and deployers should be responsible not just for transparency but for accuracy. As noted in Google’s Ambitious AI Search Changes Are Risky. Here’s Why “AI has a confidence problem. The sourcing problem in AI search is not a fringe concern. It’s been documented repeatedly, across multiple platforms, by independent researchers with no particular stake in the outcome.

A study published in 2025 by Columbia University’s Tow Center for Digital Journalism tested eight generative search tools, including Google’s Gemini, ChatGPT Search, Perplexity, and others, across 200 queries. AI search engines failed to produce accurate citations in over 60 percent of tests. Gemini and Grok were the worst performers, providing more fabricated links than correct ones. The study’s authors noted that content licensing deals with publishers, the kind AI companies have spent billions on, provided no guarantee of accurate attribution in the responses that followed.

A separate study, using the DeepTRACE evaluation framework, tested systems including Bing Chat, You.com, and Perplexity across 303 queries, scoring answers on whether claims were actually supported by the cited sources. Bing Chat had unsupported statements in 23 percent of cases. Perplexity reached 31 percent. Perplexity’s deep research agent performed worst of all, with 97.5 percent of its claims unsupported by cited sources. The researchers concluded that “current public systems fall short of their promise to deliver trustworthy, sourced responses.”

And research published jointly by the European Broadcasting Union and the BBC in October 2025 found that leading AI assistants misrepresented news content in nearly half of all responses tested, with 81 percent of responses containing some form of problem.

When asked about hallucinations at the time of the AI Overviews launch, Google CEO Sundar Pichai called them “an unsolved problem” and, in a phrase that deserves to be considered carefully, “in some ways an inherent feature” of large language models.”

Given this extent of inaccuracy it seems imperative that actors in the value chain be held responsible not just for transparency but for accuracy. It is interesting to note that that increasingly responsibility for accuracy is falling on end users rather than those who produce the underlying systems themselves as outlined in Beyond the Mirage: Beware of Generative AI and Hallucinations, and Academics in Meltdown Now That They’re Responsible for AI Hallucinations in Their Research Papers.

Question: Do existing market practices, technical tools, and legal frameworks make it easy enough to know when content is AI-generated? If not, where do gaps remain and what actions (e.g., regulatory measures, guidance and codes of conduct, standards and technical solutions, research and development, literacy initiatives, procurement requirements) do you think the Government should take?

Pauls Answer: There is a significant push to define the lack of trust in AI as a literacy issue. I think this is a false narrative. I think people are pushing back against the spread of AI into their daily lives without it being something they want. Librarians are hosting viral ‘Avoiding AI’ workshops for people who are fed up with Big Tech. There is a very real sense that AI is being forced on us. See Don’t like AI being forced on you? These librarians have the perfect first step. There is even a growing push related to anti-AI clothing. See also This shirt is like an AI invisibility cloak.

Another factor is the very real pushback against the consolidation of power related to AI in the hands of a very few high tech players. This is an equity and social issue that market practices have created. At this point in time AI technologies are not engaging in fair practices associated with reciprocity - see The Tragedy of AI. Furthermore the promise of AI to eliminate jobs, consume vast amounts of power and water to the detriment of citizens has resulted in booing of AI promoters at commencement addresses and protests against data centres leading to Canadian petitions like this. Transparency related to impact of AI on the public is swept under the carpet and replaced with ideas like watermarks. In the US Bernie Saunders introduced an AI Moratorium Act to completely freeze the construction and upgrading of new AI data centers until government passes comprehensive safeguards against job displacement and ensures these facilities do not raise utility prices for consumers.

AI has been overly hyped and is nowhere close to fulfilling the promises being made. AI has yet to even make business sense. See Is AI Profitable Yet? What will Canada’s response be to the bubble bursting? I found Bernie Saunders proposal for an AI Sovereign Wealth Fund Act arguing for public ownership of AI compelling. I like his view that:

"“It would do two extremely critical things,” Sanders said of his legislation. “First, it would give the American people a direct role in determining the future of this technology. No longer would the future of AI be dictated by a handful of Big Tech oligarchs, while the rest of the world sits back and watches them do what they want.

“Secondly, it would guarantee that the trillions of dollars potentially generated by AI are used to improve the lives of all of us — not simply to make the richest people on Earth even richer.”

And there is a very real issue related to transparency not just of the AI tools but of the entire AI technology stack (compute hardware, cloud platform, source data, enhanced data, models, API’s, apps and applications). AI relies on a suite of underlying technologies which are consolidated into the hands of a very few tech players. The consolidation of the underlying stack associated with AI in the hands of a few high tech players is largely opaque to users.

The issue isn’t just “transparency” it is also sovereignty. The Sovereign AI Index captures some of what is happening and shows Canada’s position, but it would be super helpful if Canada was more transparent around this effort and investing in a sovereign, public benefit approach to AI not a big tech one. I also think Canada should consider the sovereignty benefits associated with an open source approach to AI. The European Open Source strategy is one we too should take.

I think taking a Public AI approach will generate the greatest transparency and benefits to Canada.

As you can see the questions require quite a lot of thought and in some of the following sections my answers were considerably shorter. The next section dealt with AI Interaction

Question:When and why is it most important to know that you are interacting with an AI system? Are there contexts where you don’t need to know that you’re interacting with AI?

Pauls Answer: I always want to know I’m interacting with an AI system.

Question What factors make a disclosure of AI use meaningful and effective for the user?

Pauls Answer: AI is not human. Knowing that I’m interacting with AI affords me the opportunity to adjust my approach, ask different questions, evaluate responses differently, and adjust all aspects of interaction. Safety is a key consideration.

Question: Who in the AI value chain (developers, deployers, or others) should be responsible for providing transparency around AI interactions? Why?

Pauls Answer: Everyone along the AI value chain.

Question: Are existing market practices and legal frameworks sufficient to support transparency around AI interactions? If not, where do gaps remain and what actions (e.g., regulatory measures, guidance and codes of conduct, standards and technical solutions, research and development, literacy initiatives, procurement requirements) do you think the Government should take?

Pauls Answer: No, they are not sufficient.

The EU Artificial Intelligence Act https://artificialintelligenceact.eu/ provides one comprehensive model and frame of reference.  Japan, South Korea, and China each have their own AI Acts. All these efforts take a unique different approach aligned with national interests. All Acts and related guidance should be reviewed and evaluated for fit with Canadian needs and values.

The next section dealt with Information about AI Systems.

Question: What kinds of information about AI would you like to know in order to help you make informed choices about when and how to use AI? When and for which types of AI products is this information most important to you?

Pauls Answer: I think it will be essential for transparency and trust that Canada adopts an "open" approach to AI.

By this I don't mean a binary choice between open and closed but instead a overall approach to AI along the lines recommended by Eric Xing in "Open Weights Are Not Open Source, And It Matters Now In The Boardroom" (https://www.forbes.com/sites/patrickmoorhead/2026/09/11/open-weights-are-not-open-source-and-it-matters-now-in-the-boardroom/) who says "Our definition of openness is that you buy the house together with all the blueprints, the materials documentation, the construction details and the legal papers. If something goes wrong, you know where to fix it. If you want to remodel it, you can take it further.” To him, an API is a hotel room by comparison, and open weights is a house that comes with no blueprints.”

This article emphasizes that open is not a binary choice but instead "a ladder that gets more open the higher you climb. These are the rungs, from least to most open:

• An API you rent

• Weights under a restrictive license

• Weights under a permissive license

• Weights plus the training method

• Weights plus the data, code and checkpoints needed to rebuild the model — what some call the “open science” approach"

Canada needs a sovereign approach to AI. As the above article notes "A ministry also needs to know it can retrain the model if support for it disappears. Such a disappearance could happen because a company selling a closed model changes its commercial terms, goes out of business or gets acquired, or it could happen because a Chinese lab selling an open-weights model is steered in another direction by the Chinese government. Either way, even remote risks like this will be unacceptable to many sovereign entities.”

For it to be sovereign Canada will have to establish the entire AI stack in a way that it controls including:

• compute hardware

• cloud platform

• data inputs

• ingestion and training processes

• models

• API's

• Apps & applications

When I use AI I like to know as much about it as possible.

Openness and sovereignty are two dimensions that I think are essential for confident use.

At this stage when trust is so low I find myself only using AI under terms where my use and inputs will not be used to further train the underlying model.

In addition I'd like to know what the underlying business model of the AI is. How is it funded? Who controls it? How does that underlying model create public good?

Question: What kind of information about AI do businesses need to adopt AI safely and responsibly? Do businesses currently have access to the information that they need? How can the Government best balance the need to increase transparency (including input transparency) against the need to protect confidential business information?

Pauls Answer: Current adoption of AI is what I consider high risk. There are no safety guardrails, there is little to no transparency, the business case is vague and unproven, there is minimal trust, displacement of workers and rogue AI with the potential to threaten mankind are show stopper serious issues.

Question: Who in the AI value chain (developers, deployers, or others) is best placed to provide what information about AI systems, how they are created, and their capabilities and limitations?

Pauls Answer: Existing disclosure practices and legal frameworks are insufficient. I think all of the suggested actions will be required including regulatory measures, guidance and codes of conduct, standards and technical solutions, research and development, literacy initiatives, procurement requirements. But most of all the societal benefits of AI need to be addressed. At this point the public view is largely negative. Governments, developers, and deployers seem oblivious to these concerns.

There were many more questions some of which I didn’t answer. In the end I submitted my response to the survey questions on September 22 the day before the due date of September 23, 2026. Submissions are anonymous but it felt good to take the initiative on providing feedback.

I was especially delighted that discussions in the OEGlobal forum led to sharing of news and reactions to AI developments across Canada. Here are a few of those

• Building an AI future in Saskatchewan: Who's accountable, and who benefits?

• Bell plans to build artificial intelligence data centre near Regina, RM documents show

• Hundreds march in Vancouver to oppose planned AI data centres

• Bell plans expansion of Sask. AI data centre, making it largest private investment in province's history

• What you need to know about AI data centres coming to B.C.

• AI is our best hope for fixing Canadian healthcare

Reading these stories of actions citizens across Canada are taking in response to AI has made me super interested in learning more about how citizens across the globe are responding to AI. Citizens are concerned about water and power usage, noise, accountability, lack of consultation and input, whether AI makes economic sense, and how will the public benefit in very real ways including public or community ownership. Reading those issues makes me realize that the “Enhancing trust in AI through increased transparency” Canadian federal government consultation I’m responding to is asking a very narrow set of AI questions. Consultation with the public on these larger issues is largely absent. I appreciate the Canadian government making it possible for citizens to have a say on topics like AI transparency but I really hope these larger social issues are also brought forward for input.