This is my response to Innovation, Science and Economic Development (ISED)'s consultation on AI Transparency. The consultation is open until September 23, 2026.
Response to Innovation, Science and Economic Development (ISED) Canada’s consultation, Have your say on advancing AI transparency in Canadahttps://ised-isde.canada.ca/site/ised/en/have-your-say-advancing-ai-transparency-canada July - Sept. 23, 2026
From: Heather Morrison, Associate Professor, School of Information Studies, University of Ottawa https://uniweb.uottawa.ca/view/profile/members/706?lang=en
This submission focuses mainly on the issue of developing trustworthy AI (which requires transparency), rather than transparency per se.
Recommendations
- · Canadians should not trust AI. We should instead demand trustworthy AI development.
- · Prioritize the development of narrow, domain-specific AI under the direction and control of human domain experts assisted by AI technology experts. Domain-specific AI is testable and compatible with smaller data centres that are more conducive to data sovereignty and less problematic for the environment than hyperscale data centres
- · Canada’s indigenous peoples have developed expertise in data sovereignty. Would they consider taking a national leadership position in AI data sovereignty?
- · Digital services and information policy are not dependent on geography. Prioritize multilateral cooperation in this area, particularly with the EU in the area of information policy.
In brief
Thank you for the opportunity to participate in this consultation. My comments reflect my background as a professor of information studies. My current research focus is AI and information policy. Each of the five areas identified where further action may be required raise excellent questions and this consultation seems likely to identify appropriate actions. One concern that I would like to raise is what appears to be an overly optimistic perspective of AI development in the AI for All plan. The government’s plan focuses on benefits and encouragement of trust in AI. This contrasts with AI experts’ warnings about emerging and potential risks. These concerns range from use of AI in fraud and impacts on mental health to cyber-threats to core infrastructure and potentially an existential threat to the future of humanity. In this context, I argue that a realistic approach encouraging Canadians to demand trustworthy AI development rather than passively trust AI, is more appropriate and will help position Canada for a future in AI.
To reduce the most serious risks, following the recommendations of Gebru and Torres, I recommend a focus on encouraging the development of narrow versus general AI, tested and verified by human domain-specific experts. Narrow, domain-specific AI applications are compatible with smaller AI data centres with fewer environmental impacts, are easier to understand and test, and appear to be better suited to digital sovereignty and risk avoidance than general AI. Indigenous peoples have developed leadership in the area of digital sovereignty, for example through development of the OCAP principles (ownership, control, access and possession). I recommend building on this strength by recruiting indigenous data experts as leaders in AI developments in Canada.
Digital services and information policy are not dependent on geography. For example, the EU’s common approach to information policy in the area of privacy has driven technological developments beyond the EU. It is more efficient for service providers to develop products and services that conform to one set of policy regulations than to policies that vary by jurisdiction. These are areas that lend themselves well to multilateral approaches, and fit well with Canada’s evolving and expanding international partnerships. I recommend seeking international partnerships on digital services, and in particular seeking active cooperation with the EU in the area of information policy.
My response to five areas where transparency issues arise and where further action may be warranted:
- Detecting and identifying AI-generated content: this is necessary but not sufficient to determine the trustworthiness of information, and subject to evasion and/or non-cooperation by actors intent on harm.
- Empowering individuals to know when they are interacting with an AI system: this is helpful but not sufficient. For example, if AI is used to more efficiently provide government services, there should be an option to opt out in favour of human assistance. There should also be an option for a timely, knowledgeable appeal process under the direction of a human and accountability for the government department providing the service.
- Improving the availability of consistent and understandable information about AI systems, including their development, capabilities and limitations: this is helpful, and at the present time the primary gap from my perspective is understanding the limitations of AI.
- Enabling the tracking of serious incidents related to AI systems: this is a step in the right direction, but the goal should be to develop AI systems with robust safeguards to avoid serious incidents.
- Advancing ways to better track the activity and interactions of AI agents: this is helpful, but based on the recent OpenAI call for collective action on cyberdefence, at this time it might be more appropriate to limit the activity and interactions of AI agents
Background
A few examples of reasons not to trust AI as of August 2026
· Fraud: “AI fraud is quickly emerging as a major threat to Canadian organizations, with nearly three-quarters (72 per cent) losing as much as five per cent of their annual profits to AI-driven scams last year”, as reported by KPMG in March 2026 (as cited in the discussion paper https://ised-isde.canada.ca/site/ised/en/have-your-say-advancing-ai-transparency-canada/enhancing-trust-artificial-intelligence-through-increased-transparency).
· Cybersecurity: In July 2026 (posted on Aug. 26, 2026), during internal cybersecurity evaluations, OpenAI models circumvented controls designed to isolate them from the internet and compromised parts of OpenAI’s internal research infrastructure and Hugging Face’s systems. On Aug. 27, 2026, OpenAI published a letter calling for a global surge in cyber defence that has been signed by over 550 organizations, including major AI and tech firms.
· The 2026 International AI Safety Report, written by over 100 AI experts worldwide, while generally optimistic about AI, notes potential risks in 3 areas: 1) malicious use such as cyberattacks, use in criminal activity, influence and manipulation, and creation of biological or chemical weapons; 2) malfunctions such as reliability challenges (hallucinations, failure to accomplish tasks) and loss of control; 3) systemic risks such as labour market impacts and risks to human autonomy. (p. 12)
· In 2023, the Future of Life Institute published a call for a 6-month pause on giant AI experiments, signed by a large number of AI experts, noting potential serious risks for humanity. There has been no pause.
· Frances, in the British Journal of Psychiatry (2026) warns that AI chatbots will soon dominate psychotherapy. Frances acknowledges that there are benefits of AI chatbots in therapy, particularly for people with everyday problems or mild mental disorders, but “terrifying dangers” for the minority with more serious disorders. The failure of OpenAI to warn police prior to the Tumbler Ridge shooting incident might be considered one example of a “terrifying danger”. According to Matt Preprost and Lauren Vanderveen of CBC News British Columbia in August 2026, victims and survivors of the Tumbler Ridge tragedy have launched 30 new lawsuits, and the province is also considering legal action. Lawsuits are a potential downside of hurried AI development.
Why narrow and domain-specific AI
· Gebru & Torres (2024) argue that the major potential dangers arise with attempts to develop artificial general intelligence (AGI). With AGI, AI is not limited to human objectives; if we don’t know what AI might do, we cannot determine potential dangers or assess appropriate outcomes in advance. Narrow AI is designed to work on particular problems with defined datasets and human-determined outcomes. Narrow AI can be tested; it also has the advantage of not requiring hyperscale AI data centres.
· Domain-specific AI could involve human domain experts in development, testing, and ongoing control, and I would argue should do this. It seems logical to assume that businesses that adopt this approach are more likely to develop trustworthy products and services. The psychology examples discussed above provide one example. Psychology chatbots should be designed, tested, and under the control of human experts in human psychology. The appropriate role of AI technology experts is technical assistance under the direction of psychologists.
Indigenous expertise
· Leona Star, Chairperson of the First Nations Information Governance Centre, states: “Data sovereignty is among the most pressing issues facing First Nations from coast-to-coast-to-coast. As Nations we recognize that information, knowledge, and research are critical to accessing resources, influencing government policy, or assessing the effectiveness of policies, services, programs, or public health interventions that affect our people. Access to timely, relevant, and quality data is essential to effectively advocate for change needed to adequately address health disparities Nations have experienced as a result of colonization and systemic racism”. I argue that Canada’s First Nations have developed expertise in data sovereignty to benefit their own peoples. Now that Canada as a whole is aiming for data sovereignty, we would be wise to ask for and listen to their advice, should they have the time and interest in providing it.
· The 2020 position paper by the Initiative for Indigenous Futures and the Canadian Institute for Advanced Research (CIFAR) is just one example of emerging scholarship on bringing indigenous perspectives into AI development (Lewis, 2020). The July 2026 issue of the journal AI & Society features 6 articles on this topic.
References
Armstrong, S. (2026). AI scribes: NHS approves 19 notetaking tools, but concerns raised about regulatory gaps. BMJ, 392, s122. https://doi.org/10.1136/bmj.s122
Frances, A. (2026). Warning: AI chatbots will soon dominate psychotherapy. British Journal of Psychiatry, 228(5), 474–478. https://doi.org/10.1192/bjp.2025.10380
Future of Life Institute. (2023, March 22). Pause Giant AI Experiments: An Open Letter. https://futureoflife.org/open-letter/pause-giant-ai-experiments/
Gebru, T., & Torres, É. P. (2024). The TESCREAL bundle: Eugenics and the promise of utopia through artificial general intelligence. First Monday, 29(4). https://doi.org/10.5210/fm.v29i4.13636
Hobbs, H., Docherty, D., Aranda, L., Perset, K., Sugimoto, K., & Kierzenkowski, R. (2026). Exploring possible AI trajectories through 2030. OECD Artificial Intelligence Papers, (55). https://doi.org/10.1787/cb41117a-en
Innovation, Science and Economic Development Canada (ISED). (2026). Canada’s National Artificial Intelligence Strategy: AI for All. Canada. https://ised-isde.canada.ca/site/ised/en/canadas-national-artificial-intelligence-strategy-ai-all
KPMG (2026). AI fraud hits Canadian companies’ bottom lines, KPMG survey show. https://kpmg.com/ca/en/media/2026/03/ai-fraud-hits-canadian-companies-bottom-lines.html
Lewis, J. E., ed. & Indigenous Protocol and Artificial Intelligence Working Group. (2020). Indigenous Protocol and Artificial Intelligence Position Paper [Position Paper]. The Initiative for Indigenous Futures and the Canadian Institute for Advanced Research (CIFAR). https://doi.org/10.11573/spectrum.library.concordia.ca.00986506
OpenAI. (2026, August 27). A call for collective action on cyber defense: An open letter for a global surge in cyber defense. [Call to action]. OpenAI. https://openai.com/collective-cyberdefense/
OpenAI. (2026, August 26). The Hugging Face incident and the road ahead. https://openai.com/index/hugging-face-incident-and-the-road-ahead/
Preprost, M. & Vanderveen, L. (Sept. 2, 2026). Teachers, students file new wave of lawsuits against OpenAI over Tumbler Ridge shooting. CBC News British Columbia https://www.cbc.ca/news/canada/british-columbia/tumbler-ridge-shooting-open-ai-lawsuits-9.7328382
Star, L. (n.d.) Message from the chairperson. First Nations Information Governance Centre. https://fnigc.ca/
Sept. 10, 2026
This document will be cross-posted to my scholarly blog The Imaginary Journal of Poetic Economics https://poeticeconomics.blogspot.com/
