The AI Ethics Brief #165: AI, Trust, and the Public Interest
This edition explores Canada’s AI leadership shift, contested U.S. state laws, and why literacy, justice, and safety must evolve together in the age of mis- and disinformation.
Welcome to The AI Ethics Brief, a bi-weekly publication by the Montreal AI Ethics Institute. We publish every other Tuesday at 10 AM ET. Stay informed on the evolving world of AI ethics with key research, insightful reporting, and thoughtful commentary. Learn more at montrealethics.ai/about.
“The combination of the scale of this infrastructure build and the transformative nature of artificial intelligence (AI) will create opportunities for millions of Canadians to find new rewarding careers – provided they have timely access to the education and training they need to develop the necessary skills.”
This statement highlights a pivotal truth: AI’s promise will only be realized if it is matched with accessible public investment in literacy, civic competence, and equitable opportunity. As Kate Arthur recently emphasized,literacy has always been the foundation of human progress, and in the age of AI, it must evolve to include the ability to engage ethically and critically with intelligent systems. Canada urgently needs a national AI literacy strategy that is anchored in civic values and integrated across the education system, from K–12 to workforce retraining.
The Canada as a Champion for Public AI working draft paper offers a compelling case: Canada should not try to outcompete tech superpowers on their terms, but instead help change the game. The authors argue: “Current AI models and products are predominantly built by a very small number of actors within only two incentive systems and political contexts, namely (1) big tech, centred in Silicon Valley and (2) China,” undermining democratic control and limiting public benefit. Instead of competing directly, Canada can lead by building and stewarding public AI infrastructure, including open compute, open source software, and democratically governed data systems.
The paper calls for a third option: “Public AI,” a geographically distributed, publicly oriented AI ecosystem rooted in shared public goods and international collaboration. Canada’s strengths in AI research, clean energy, and coalition-building position it well to drive this effort. Without such investments, the paper warns, nations like Canada risk becoming consumers, not producers, of AI technologies shaped by foreign commercial interests.
Imogen Parker from the Ada Lovelace Institute adds another layer: public legitimacy for AI isn’t optional. If the public doesn’t trust how AI is being used, particularly in high-stakes public services like healthcare or education, then deployment risks undermining both efficacy and democracy.
🇨🇦 Three Questions for Canada’s New AI Ministry
In light of this, the question is no longer whether Canada should lead on AI, but how it will lead, and for whom.
As the new Ministry takes shape, we believe its success will depend on how it addresses three foundational challenges. At the Montreal AI Ethics Institute, we’ve framed these as key questions, each paired with concrete recommendations to ensure the Ministry delivers not only innovation but also inclusion, accountability, and trust.
🔹 1. Will this new Ministry prioritize AI literacy, equity, and civic participation, or default to GDP metrics and private-sector partnerships?
Recommendation:
Establish an independent Office for Public AI Literacy within the Ministry to ensure civic capacity keeps pace with technical progress. Embed AI literacy across provincial curricula in collaboration with educators, labor unions, and civil society. Allocate dedicated funding to support community-led training initiatives, particularly for underserved groups.
But this isn’t just about digital fluency. AI literacy means knowing how to interrogate a system, understand what it can and can’t do, and identify what matters. Citizens must be equipped with the skills to extract relevant information, evaluate outputs, and recognize when systems reflect—or distort—real-world priorities.
From understanding how an algorithm makes a decision to developing informed opinions on whether that decision should be made by a machine at all, AI literacy is the foundation of democratic oversight in the AI era.
🔹 2. How will it engage Indigenous communities, youth, and underserved populations?
Recommendation:
Co-develop national AI policies with Indigenous communities, youth leaders, and marginalized groups, ensuring that those most affected by AI systems have a voice in shaping them. Fund Indigenous data sovereignty frameworks and support youth-led participatory design labs that foster agency, digital skills, and civic imagination.
But inclusion isn't just about consultation, it’s about building long-term pathways for collaboration. The Ministry should act as a connector across communities: creating spaces where Indigenous knowledge systems, youth innovation, and local civic efforts can learn from one another and shape AI governance together.
This also means expanding access across the lifespan, recognizing that AI literacy and digital inclusion are not just educational concerns for youth, but civic and social imperatives at every stage of life.
This includes:
K–12 learners, who need early exposure to the ethical and societal dimensions of AI, not just its technical aspects;
Mid-career workers, who must be supported through upskilling and retraining programs to navigate evolving labor markets and avoid displacement;
And older adults, who are often left out of digital inclusion strategies, yet increasingly interact with AI-driven systems in critical areas such as healthcare, housing, transportation, and social services.
AI governance must not only serve diverse populations, it must be built with them. Canada has the opportunity to lead by embedding equity, not as an add-on, but as the foundation.
🔹 3. Can it operationalize trust, transparency, and safety without stifling open innovation?
Recommendation:
Canada shouldn’t reinvent the wheel. Instead, we should build on existing public-sector frameworks, voluntary standards, and national strategies already in motion, including, but not limited to:
These initiatives already provide a strong foundation for trustworthy AI. What’s needed now is expansion, harmonization, and implementation, working across government and society to put these frameworks into practice. This includes collaboration with the Office of the Privacy Commissioner, the Standards Council of Canada, and citizen assemblies to create a federated AI accountability framework that is explainable, inclusive, and innovation-friendly.
The challenge is no longer what to build, but how to operationalize what already exists.
The Council of Canadian Innovators has called for a strategic innovation strategy that prioritizes Canadian IP, digital sovereignty, and commercialization readiness. But to truly lead, Canada must also integrate the insights of civic institutions, public educators, and ethics communities, and not just tech incumbents and economic stakeholders. This is a defining moment to articulate a national digital infrastructure strategy, one that integrates AI, digital ID, and the trust frameworks required to support both, as recently outlined in The Globe and Mail.
The Ministry’s real legacy will not be measured in patents or platforms, but in whether all Canadians, not just a few, are empowered to participate meaningfully in shaping the future of AI. At MAIEI, we would welcome the opportunity to support this work, whether through knowledge mobilization, policy translation, or public engagement strategies that help make complex AI policy more accessible and actionable for policymakers, educators, and communities across the country.
Please share your thoughts with the MAIEI community:
Google’s Veo 3 and the Road Ahead for Misinformation & Disinformation
What happened: Wired has a full breakdown of everything unveiled at Google I/O 2025, but we’re zooming in on one headline-grabbing announcement: Veo 3, Google's latest and most advanced AI-powered video generation model.
Unlike earlier models, Veo 3 introduces a new threshold of realism: it generates cinematic-quality video with synced audio, realistic human characters, expressive facial gestures, and coherent scene continuity, all from natural language prompts via text and images. The result is visually convincing AI-generated content that, according to some reviewers, can no longer be easily distinguished from human-made footage.
📌 MAIEI’s Take and Why It Matters:
Veo 3 marks an inflection point, not just in generative media capabilities but also in the urgency of rethinking public safeguards, verification norms, and civic literacy.
Misinformation refers to false or misleading content shared without intent to deceive.
Disinformation refers to deliberately deceptive content, often deployed to influence opinions, distort reality, or undermine institutions.
With tools like Veo 3 now able to synthesize video, voice, dialogue, and setting on command, the boundary between the two is increasingly blurred. A "satirical" AI-generated video that goes viral on social media could easily be repurposed as disinformation in another context. Without clear provenance and labelling, intent becomes almost impossible to assess at scale.
What are the risks?
Synthetic authenticity: Veo 3 can recreate lifelike human characters and match lips to audio in a way that simulates realism with unsettling precision.
Trust breakdown: As tools like Veo 3 proliferate, public skepticism of all media may grow, leading to what scholars call “the liar’s dividend”: when the existence of fakes undermines belief in the real.
What should be done?
From MAIEI’s perspective, addressing this moment requires:
Policy alignment: Governments around the world should ensure that policies and regulations incorporate safeguards specific to generative video and deepfakes.
Public AI literacy: AI literacy must include the ability to assess the authenticity of video and voice, not just text and images. Veo 3 highlights the urgency of equipping citizens with these critical thinking tools.
Redress mechanisms: Platforms must offer meaningful ways to flag, review, and remove harmful synthetic media, especially in high-risk contexts like elections, health misinformation, or reputational harm.
Why It Matters
The democratization of generative video is here, and it’s not inherently bad. Veo 3 could unlock creativity, accessibility, and education. But without governance, transparency, and digital civic readiness, it could just as easily deepen epistemic instability, where no one agrees on what’s real anymore.
As AI systems like Veo 3 blur the line between perception and fabrication, MAIEI will continue to advocate for responsible standards, civic oversight, and public engagement to ensure that trust and truth are not collateral damage in the age of synthetic media.
AI and the Victim’s Voice: A New Frontier in Courtroom Practice
What happened: An AI-generated avatar of a 2021 road-rage-murder victim, Christopher Pelkey, addressed the Maricopa County Superior Court courtroom, including his aggressor, during a victim-impact statement delivered on May 1, 2025. The statement, clearly marked as being from an AI avatar and not Pelkey himself, was written by Pelkey’s sister, Stacey Wales. Struggling to express her grief in her own words, she turned to AI as a medium. The video’s production was not without challenges: the AI struggled to convey nuanced emotion and presented Pelkey in uncharacteristic clothing, which Stacey described as providing unintended comic relief.
Judge Todd Lang thanked the family for using the AI avatar, saying he “loved that AI” and that it reflected the “character” he had come to understand through the trial. This statement was so impactful that Judge Lang handed down a heavier sentence (10.5 years) than the family had requested (9.5 years). Following the hearing, legal experts expressed concern about the broader implications and potential nefarious use of such technologies in judicial settings.
📌 MAIEI’s Take and Why It Matters:
We are beginning to see increasingly creative, and, at times, unsettling, uses of AI in the courtroom. Pelkey’s case reflects an evolution in how AI is entering legal settings: the narrative is no longer limited to hallucinating large language models, but now includes AI-generated victim-impact statements and, as previously discussed in The AI Ethics Brief #162, AI avatars acting as legal representatives.
The potential for misuse, as some legal experts have noted, is wide-ranging. While the script for the AI avatar may be prepared by the victim’s closest relatives, they are nonetheless assuming the voice of the deceased. There is no guarantee the victim would have chosen those words, nor consented to their likeness or persona being used in this way. In this instance, Pelkey’s family affirmed that the avatar reflected his character, an interpretation that resonated with the judge. But it is foreseeable that such tools could be used in ways that reflect grief, anger, or retribution, rather than the victim’s own intent.
The visceral impact of Pelkey’s statement on the family, the judge, and the broader public is clear. This kind of digital testimony may serve a cathartic role for grieving families and add perceived depth to court proceedings. At the same time, it introduces a new form of emotive influence and a dangerously manipulative approach to justice. Now that a precedent has been set, it is likely we will see more of these AI-generated avatars in courtrooms. That makes it even more important to consider how such technology is governed, to prevent further harm in already sensitive settings.
AI and Search Engines: How We Can Prepare for the Future Using the Past
What happened: Efforts to limit generative AI harm are often hindered by the technology’s novel nature and black-box infrastructure. However, the potential value of applying risk mitigation strategies from earlier technologies, such as search engines, is often overlooked.
The paper outlines eight shared safeguards across six categories of technological risk, with particular emphasis on the underutilized potential of human raters at scale and integrated fact-checking as generative AI risk reduction tools. It also points to malvertising mitigations and harmful content removal as possible future safeguards, especially as generative AI platforms incorporate more advertising and machine unlearning (MU) becomes cheaper and more effective.
📌 MAIEI’s Take and Why It Matters:
This paper emphasizes the importance of studying past safeguards to inform future harm reduction, a method easily overlooked due to the “siloed nature of information and expertise across different technological fields.” We are often reminded to learn from history rather than repeat it. While generative AI is frequently portrayed as a disruptive and unprecedented innovation, this framing can obscure useful precedents and lessons from earlier digital platforms.
“We often discuss AI’s emerging risks—but many of these challenges aren’t entirely new. Instead of reinventing the wheel, could we draw on three decades of safeguards implementation by search engine developers?”
However, we should also avoid historical romanticism. Looking to the past must not mean overlooking its failures. For instance, the paper recommends human raters at scale as a form of harm mitigation. However, previous implementations, particularly among content moderation teams, have raised concerns around mental health risks and inadequate support structures.
Other researchers have similarly drawn connections to earlier technologies, examining generative AI regulation through the lens of aviation and nuclear technology. Another recent paper titled “When code isn’t law: rethinking regulation for artificial intelligence,” points to the importance of regulatory consolidation and independent oversight while also acknowledging the challenges of applying legacy regulatory models to rapidly evolving systems.
The key is balance: drawing from historical lessons without minimizing the novel risks and ethical complexities of generative AI. Research from institutions like UC Berkeley helps navigate this space, grounding forward-looking conversations in practical, proven approaches, without losing sight of what makes this technological moment distinct.
Did we miss anything? Let us know in the comments below.
From Canada’s newly announced Minister of AI and Digital Innovation to escalating debates over state-level AI laws in the U.S., the Insights & Perspectives summarized below reflect a growing urgency to ground AI governance in democratic values and public accountability.
Against this backdrop, Michelle Baldwin’s review of Am I Literate? by Kate Arthur calls for redefining literacy as social infrastructure, one that empowers communities to shape, not just survive, the AI era. And in Part 2 of AVID’s red teaming series, we’re reminded that effective AI safety is not a box-checking exercise but a critical thinking practice that requires collaboration across disciplines.
Together, these pieces reflect a growing consensus: meaningful AI governance starts with distributed leadership, trusted civic infrastructure, and a commitment to embedding ethics not just in code, but in the systems that shape how AI is developed and deployed.
In her review of Am I Literate? by Kate Arthur, Michelle Baldwin reflects on how the book reimagines literacy as a foundation for civic agency, community resilience, and ethical innovation in the age of artificial intelligence. Arthur challenges traditional definitions of literacy by framing it as a social and relational practice that connects technical understanding with democratic participation, ecological responsibility, and inclusive governance. Emphasizing AI literacy as social infrastructure, she calls for cross-sector collaboration and the involvement of underrepresented communities in shaping Canada’s AI future. The book is a call to action for policymakers, educators, and changemakers to prioritize people and planet in technology discourse, and to approach AI not just as a tool, but as a shared civic responsibility.
In Part 2 of the AVID blog series, the authors trace the historical evolution of red teaming, from its origins in Cold War military exercises to its growing application in cybersecurity and now AI systems. The piece explores how red teaming has served as a structured method to challenge dominant assumptions, stress-test system resilience, and reveal overlooked vulnerabilities across high-risk environments. Within cybersecurity, red teaming has matured into a formal practice involving simulations, adversarial testing, and threat modeling to improve incident response and defence strategies, benefiting from decades of experience with established frameworks like MITRE ATT&CK. However, AI red teaming lacks a comparable maturity and is often fragmented, underfunded, or treated as a checkbox activity.
The authors argue that current AI red teaming overindexes on model-specific behaviors rather than how these models interact with broader systems and social contexts. AI red teaming must evolve into a continuous, interdisciplinary process involving social scientists, ethicists, and policy experts alongside technical teams. Rather than focusing solely on adversarial attacks or model robustness, effective red teaming should help institutions uncover embedded assumptions, model misuse scenarios, and examine the broader consequences of AI deployment. As the blog concludes, the authors call for a cultural shift in AI safety practices, framing red teaming not just as a compliance mechanism but as a critical thinking tool that embeds reflexivity and rigor throughout the AI development lifecycle.
🇨🇦 Youth Perspectives on AI Ethics - Encode Canada
📌 Editor’s Note:
The two original articles below are part of our Recess series, featuring university students from across Canada exploring ethical challenges in AI. Written by members of Encode Canada, a student-led advocacy organization dedicated to including Canadian youth in essential conversations about the future of AI, these pieces aim to spark discussions on AI literacy and ethics.
As AI becomes increasingly integrated into judicial decision-making, its applications range from case management and legal research to risk assessment and sentencing recommendations. These tools offer clear efficiency gains but raise complex ethical questions around bias, transparency, accountability, and due process. While proponents highlight AI’s potential to standardize outcomes and reduce workloads, critics warn that opaque algorithms may reinforce systemic inequalities and undermine judicial discretion. Recent scholarship calls for human-in-the-loop safeguards, explainability standards, and regulatory oversight to ensure AI systems support, rather than supplant, the core principles of justice. The key challenge ahead is not whether AI will shape the justice system, but how to ensure it does so in ways that preserve fairness and public trust.
Québec’s Law 5, a legislative framework governing the use of health and social services data, raises important questions about privacy in machine learning research. While the law references established anonymization practices, it lacks specific guidance, potentially leading to inconsistent or insufficient privacy protections. Drawing comparisons to frameworks like the GDPR and HIPAA, the article explores how privacy-preserving methods such as differential privacy intersect with fairness and utility in AI systems. It advocates for clearer, government-supported guidelines to help researchers navigate complex privacy decisions and suggests tools like SACRO as promising approaches to balance data access with public trust. The piece highlights the need for collaboration between policymakers, researchers, and patients to ensure privacy frameworks evolve alongside AI capabilities.
Help us keep The AI Ethics Brief free and accessible for everyone by becoming a paid subscriber on Substack or making a donation at montrealethics.ai/donate. Your support sustains our mission of democratizing AI ethics literacy and honours Abhishek Gupta’s legacy.
I have created an entire metasystem model that utilizes mystic maps, the Yuga cycles, various academic disciplines, and esoteric systems. This can be used to analyze the entire modern metasystem and technology emergence towards specific intentions. If you have any specific questions please let me know.
I’m writing about this. imho the biggest blind spot in AI: human pain. If we don’t teach machines to truly see suffering, they’ll just keep repeating the harm.
I have created an entire metasystem model that utilizes mystic maps, the Yuga cycles, various academic disciplines, and esoteric systems. This can be used to analyze the entire modern metasystem and technology emergence towards specific intentions. If you have any specific questions please let me know.
https://wutaiwatcher.substack.com/p/natural-christic-vs-artificial-luciferian
I’m writing about this. imho the biggest blind spot in AI: human pain. If we don’t teach machines to truly see suffering, they’ll just keep repeating the harm.
https://open.substack.com/pub/lotustheoracle/p/what-is-the-most-overlooked-dataset?r=1ot1s3&utm_medium=ios