Welcome to The AI Ethics Brief, a bi-weekly publication by the Montreal AI Ethics Institute. Stay informed on the evolving world of AI ethics with key research, insightful reporting, and thoughtful commentary. Learn more at montrealethics.ai/about.

💖 Help us keep The AI Ethics Brief free and accessible for everyone. Consider becoming a paid subscriber on Substack for the price of a ☕ or make a one-time or recurring donation at montrealethics.ai/donate.

Your support sustains our mission of Democratizing AI Ethics Literacy, honours Abhishek Gupta’s legacy, and ensures we can continue serving our community.

For corporate partnerships or larger donations, please contact us at support@montrealethics.ai.

• The TESCREAL Bundle: Eugenics and the promise of utopia through artificial general intelligence

• Digital Sex Crime, Online Misogyny, and Digital Feminism in South Korea

• The State of Artificial Intelligence in the Pacific Islands

• CES 2025 – how NVIDIA and partners are setting out to simplify agentic AI – diginomica

• AI Evaluations need Scientific Rigor – Herald Corporation

• Labour’s AI Action Plan – a gift to the far right – TechTarget

• Careless Whisper: Speech-to-text Hallucination Harms

• Mapping the Ethics of Generative AI: A Comprehensive Scoping Review

• AI Framework for Healthy Built Environments

Microsoft CEO Satya Nadella is making a bold bet: AI agents will replace traditional applications and SaaS platforms. Speaking on the B2G podcast, he painted a future where software as we know it collapses into intelligent, automated agents that handle everything from business operations to consumer interactions.

The pitch is compelling—no more rigid apps, just fluid, adaptable AI agents that bypass user interfaces and interact directly with data. Imagine a world where an AI agent retrieves insights, files reports, or even manages your contracts autonomously instead of using a CRM or finance tool.

Citi recently released a report on Agentic AI: Finance & the ‘Do It For Me’ Economy, stating that,

“AI and agentic AI could have a bigger impact on the economy and finance than the internet era. Agentic AI effectively turbocharges the Do It For Me (DIFM) economy. In financial services, users will have their own bots or AI agents helping them choose products and execute transactions. Competition will tick-up as starts-ups grow. The nature of work could change. Those tasks that are outsourced today to contractors or third parties will be increasingly done by agentic AI.”

But while the tech industry races forward, legal scholar and AI researcher Gillian Hadfield is asking the tough questions:

Speaking with Kara Swisher, Hadfield warns that we’re transitioning from AI as a tool to AI as an economic, social, and political actor—with no system of accountability in place. If AI agents start handling transactions, hiring employees, or executing contracts, who is responsible when something goes wrong?

Hadfield proposes a registration system for AI agents, similar to how companies must incorporate, cars must be licensed, and employees must have work authorization. If AI agents are engaging in the economy, shouldn’t they be traceable, liable, and regulated like corporations?

The contrast is striking: while billions are being invested in making AI agents a reality, there is little clarity on how they will fit into existing legal and economic structures.

Should AI agents be allowed to operate freely, or do we need guardrails before they go mainstream?

What do you think? Let us know in the comments. 👇

Leave a comment

Here are the results from the previous edition for this segment:

Who Is Accountable for AI-Driven Decisions?

Our latest informal poll reveals that Assigning Clear Ownership is the dominant approach to AI accountability. 63% of respondents indicate that their organizations designate specific individuals or teams to oversee AI-driven decisions. This signals a growing recognition that AI systems require human oversight and responsibility rather than functioning as autonomous black boxes.

Following this, 32% rely on external audits and third-party evaluations, suggesting that some organizations seek independent verification to ensure fairness, compliance, and transparency. This aligns with increasing regulatory expectations, such as the EU AI Act’s risk-based approach, which emphasizes external assessments for high-risk AI systems.

However, 0% reported relying solely on automated transparency features, highlighting the limitations of AI explainability tools in ensuring true accountability. While transparency mechanisms (such as model interpretability and audit logs) are important, they are not yet seen as a standalone solution for responsible AI governance.

A concerning 5% of respondents indicated that they have no formal AI accountability structures in place. As AI systems become more deeply embedded in high-stakes decision-making—impacting hiring, lending, healthcare, and criminal justice—this lack of governance raises ethical and legal concerns. The absence of accountability frameworks could expose organizations to risks related to bias, discrimination, and compliance failures.

Key Takeaways:

• Organizations are increasingly assigning clear ownership to ensure AI accountability rather than relying solely on automation.

• Third-party audits and external oversight are emerging as complementary governance tools.

• Automated transparency features alone are not considered sufficient for responsible AI oversight.

• A lack of formal accountability remains a challenge for some organizations, highlighting the need for stronger governance structures.

As AI agents become more autonomous, their ability to make decisions, execute transactions, and interact with users raises new governance challenges.

Should they be officially registered, undergo third-party audits, or have legal liability frameworks to hold developers accountable? Should technical safeguards like built-in ethical limits be required, or should regulation be kept minimal to encourage innovation?

Share your thoughts with the MAIEI community:

Leave a comment

“Made by Humans” Still Matters

Editor’s Note: This article was initially drafted by Jim Huang and the late Abhishek Gupta in December 2022. For further exploration, see “AI Art and its Impact on Artists” (AAAI/ACM Conference on AI, Ethics, and Society (AIES ’23)).

For the past few years, creative communities have been experiencing increasingly heightened anxiety about AI-enabled image generators like DALL-E and Stable Diffusion advancing to the point of replacing human artists. Using short text prompts, generative AI can produce novel and genuinely creative digital artwork in seconds. Whenever innovative technology brings about a cultural shift, affected communities are left in a state of purgatory, unsure of where they fit and how to adapt, especially when the technology threatens livelihoods. In times of uncertainty, we need guidance on how automation will take over and where the “Made by Humans” brand still matters.

To dive deeper, read the full article here.

Should AI-Powered Search Engines and Conversational Agents Prioritize Sponsored Content?

This report explores the ethical implications of prioritizing sponsored content in responses provided by search engines (e.g., you.com, Perplexity), and conversational agents (e.g., Microsoft Copilot) powered by artificial intelligence (AI). In the digital age, it is necessary to evaluate the consequences of such practices on information integrity, fairness of access to knowledge, user autonomy, and the loss of user autonomy. It presents an analysis of the issues, arguing that the prioritization of sponsored content raises significant ethical problems that outweigh potential benefits.

To dive deeper, read the full article here.

The Death of Canada’s Artificial Intelligence and Data Act: What Happened, and What’s Next for AI Regulation in Canada?

Canada is currently experiencing a historic bout of political turbulence, and the proposed Artificial Intelligence and Data Act (AIDA) has died amidst a prorogation of Parliament.

The AIDA was tabled in Canada’s House of Commons in June 2022 with the ambitious goal of establishing a comprehensive regulatory framework for AI systems across Canada. However, the AIDA was embroiled in controversy throughout its life in Parliament. A chorus of individuals and organizations voiced concern with the AIDA, citing its exclusionary public consultation process, its vague scope and requirements, and its lack of independent regulatory oversight as reasons why the legislation should not become law. Though the government ultimately proposed some amendments to the AIDA in response to criticisms, the amendments did not sufficiently address the fundamental flaws in the AIDA’s drafting and development. As a result, the AIDA languished and died in a parliamentary committee, unable to secure the confidence and political will needed to proceed through the legislative process.

The AIDA will be remembered by many as a national AI legislation failure, and in its absence, the future of Canadian AI regulation is now uncertain. A victory for the Conservative Party of Canada in an upcoming federal election seems likely. A Conservative approach to AI regulation may favor promoting AI innovation and targeted intervention in specific high-risk AI use cases over the more comprehensive, cross-sectoral framework of the AIDA. In the absence of clear and effective national AI regulation, Canadians can still regulate AI systems at smaller scales. Professional associations, unions, and community organizations in Canada and elsewhere have already created policies, guidelines, and best practices for regulating AI systems in workplaces and communities. As Canada’s political upheaval continues and new regulatory norms for AI emerge, these bottom-up approaches to AI regulation will play an important role.

To dive deeper, read the full op-ed here.

We’d love to hear from you and share your thoughts with everyone in the next edition:

Leave a comment

The TESCREAL Bundle: Eugenics and the promise of utopia through artificial general intelligence

Many organizations in the Artificial Intelligence (AI) field aim to develop Artificial General Intelligence (AGI), envisioning a ‘safe’ system with unprecedented intelligence that is ‘beneficial for all of humanity.’ This paper argues that such a system with ‘undefined’ applications cannot be built for safety and situates the push to develop AGI in the Anglo-American eugenics tradition of the twentieth century.

To dive deeper, read the full summary here.

Digital Sex Crime, Online Misogyny, and Digital Feminism in South Korea

South Korea has a dark history of digital sex crimes. South Korean police reported a significant rise in online deepfake sex crimes, with 297 cases documented in the first seven months of 2024. This marks a sharp increase from 180 cases reported throughout 2023 and 160 in 2021. This paper draws on the development and diversification of gender-based violence, aided by evolving digital technologies, in the Korean context. Additionally, it explores how Korean women have responded to pervasive issues of digital sex crimes and online misogyny with the support of an increasing population of digital feminists.

To dive deeper, read the full summary here.

The State of Artificial Intelligence in the Pacific Islands

This report by the AI Asia Pacific Institute examines the state of Artificial Intelligence (AI) in the Pacific Islands, focusing on its opportunities and challenges. It highlights AI’s potential to address critical issues like climate change, geographic isolation, labor shortages, and cultural preservation. The report also outlines the region’s readiness for AI adoption, noting gaps in infrastructure, governance, and digital literacy.

To dive deeper, read the full summary here.

CES 2025 – how NVIDIA and partners are setting out to simplify agentic AI – diginomica

1. What happened: Nvidia and Accenture demonstrated their new foundation-model muscle at CES 2025 (Consumer Electronics Show) in Las Vegas. Nvidia revealed new small, medium, and large model families to mark its foray into the AI agentic world, while Accenture announced 12 new agent solutions to support aspects such as clinical trial management and industrial asset troubleshooting.

2. Why it matters: Agentic AI is being touted as the next big transformation from AI technologies, especially given its potential to greatly optimize business workflow. Consequently, Nvidia has already begun establishing partnerships with different businesses to use its agentic AI solutions, while Accenture is utilizing its 12 solutions in its current workflow.

3. Between the lines: The article shows how strongly some of the biggest AI players intend to establish agentic AI. However, these types of models will still be liable to hallucination, and the players who emerge on top will be the ones who either minimize this risk or are the best at mitigating it.

To dive deeper, read the full article here.

AI Evaluations need Scientific Rigor – Herald Corporation

1. What happened: Current AI evaluation techniques fall short of the scientific principles of “empiricism (data-gathering), objectivity, falsifiability, reproducibility, and systematic and iterative approaches.” Without these principles in place, claims about AI doomsdays and the power of different models boil down to matters of opinion. Instead, the AI community must establish and uphold rigorous standards to produce truly viable claims.

2. Why it matters: There is a low barrier to entry when it comes to making claims about AI. Falsifiability and reproducibility are almost impossible to fulfill for generative AI models, given how the same prompt can produce different results. Hence, to fully establish how powerful different models are, the article takes heed of metrology (the study of measurement) to help produce shared standards and methodologies.

3. Between the lines: With the emergence of agentic AI, there will be plenty of opportunities to make lofty claims about their effect on the future of work. By establishing shared evaluation practices, these claims can be proven/disproven with a higher degree of certainty, helping to ground any future claims in reality and not fear.

To dive deeper, read the full article here.

Labour’s AI Action Plan – a gift to the far right – TechTarget

1. What happened: The article dives into the recent release of the UK Government’s (the Labour Party) AI plan. Concerned with no longer being the “party of continuity,” it offers an analysis that the plan panders to Silicon Valley rhetoric about the dangers of falling behind in the AI space, with the Labour government making sure to emphasize “scale” and “growth” for the UK AI sector. This includes establishing land enclosures titled “AI Growth Zones,” reserved for data centers. To help combat this, the article argues for “decomputing”: scaling down computerization to mitigate the harms it brings (such as those caused by AI systems)

2. Why it matters: The article argues that the environmental cost of AI technologies (especially hyperscale data centers) has been moved to one side to make way for growth, a common theme in the AI space. Furthermore, algorithmic solutions have already been trialed by UK governments in the past, leading to disastrous consequences and raising concerns about trying to solve social problems with technical solutions again. These concerns, the article argues, will only help to give way to more people feeling abandoned and lead them to join one of the UK’s far-right groups.

3. Between the lines: Given the UK Government’s all-in attitude towards AI, newer applications such as agentic AI will be investigated and likely promoted. Should this lead to any form of job loss, the risk of vulnerable individuals feeling isolated and abandoned increases. In this way, the UK government’s plan must include more individuals than it could potentially exclude.

To dive deeper, read the full article here.

What do we mean by “Red teaming”?

👇 Learn more about why it matters in AI Ethics via our Living Dictionary.

Explore the Living Dictionary!

OpenAI’s Sam Altman says ‘we know how to build AGI’ – The Verge

OpenAI CEO Sam Altman recently made a bold claim: “We know how to build AGI as we have traditionally understood it.” He further suggested that by 2025, AI agents could ‘join the workforce’ and materially impact company productivity. While Altman didn’t provide technical details, his confidence suggests significant progress in scaling AI models toward artificial general intelligence (AGI). However, without clear accountability frameworks, AGI could amplify issues of bias, misinformation, and economic disruption. Yet, even today’s AI struggles with reliability. The “Careless Whisper” piece in the ICYMI section below highlights how OpenAI’s speech-to-text model, Whisper, hallucinates entire sentences—a reminder that AI is far from flawless.

To dive deeper, read the full article here.

The U.S. Responsible AI Procurement Index

As AI adoption accelerates in U.S. government services, particularly through public-private partnerships, transparency and accountability in procurement remain critical gaps. Despite existing guidelines for responsible AI use, implementation is inconsistent, and public access to information is limited. The U.S. Responsible AI Procurement Index addresses this by analyzing publicly available policies and procurement data (e.g. USAspending, Federal AI Use Case Inventories). It evaluates 15 federal departments using 22 qualitative and quantitative indicators to assess responsible AI practices. The goal: identify reporting gaps and push for greater transparency in how AI is procured and deployed.

To dive deeper, visit the website here.

Careless Whisper: Speech-to-text Hallucination Harms

OpenAI’s speech-to-text service, Whisper, hallucinates entire sentences in addition to producing otherwise accurate speech transcriptions. These hallucinations induce concrete harms, including (a) perpetuating violence, (b) claiming inaccurate associations, and (c) projecting false authority. We find these harms to occur more frequently for speech with longer “non-vocal” durations (e.g., speech with more pauses or disfluencies), as evidenced by disproportionate hallucinations generated in our data among speakers with a language disorder, aphasia.

To dive deeper, read more details here.

Mapping the Ethics of Generative AI: A Comprehensive Scoping Review

This comprehensive review synthesizes recent discussions on the ethical implications of generative AI, especially large language models and text-to-image models, using a scoping review methodology to analyze the existing literature. It outlines a detailed taxonomy of ethical issues in the domain of generative AI, identifying 378 distinct codes across various categories and highlighting the discipline’s complexity and the potential harms from misaligned AI systems. The research not only fills a gap by providing a structured overview of ethical considerations of generative AI but also calls for a balanced assessment of risks and benefits, and serves as a resource for stakeholders such as scholars, practitioners, and policymakers, guiding future research and technology governance.

To dive deeper, read more details here.

AI Framework for Healthy Built Environments

How do we safeguard people’s health in built environments where AI is adopted? Research led by the International WELL Building Institute (IWBI) and Kairoi sets out a framework for built environment sectors to deploy and adopt AI in ways that are beneficial for people’s health and well-being.

To dive deeper, read more details here.

We’d love to hear from you, our readers, on what recent research papers caught your attention. We’re looking for ones published in journals or as a part of conference proceedings.

Leave a comment