Toward a Canadian transnational diplomatic strategy for the regulation of AI-decision support systems

The integration of artificial intelligence (AI) into the defence sector has been growing since 2020. In response, the Department of Defense of Canada published its “Strategy on artificial intelligence” [i] in 2024. As AI becomes increasingly integrated into defence capabilities, it is essential for Canada to assess the risks involved and work in collaboration with civil society, legal experts and researchers to develop a legally binding instrument within relevant international forums.

The use of AI in military decision-making is one of the most pressing contemporary issues in disarmament: its use has significant implications for human security and dignity, human rights, and gender equality. To this end, and in accordance with Canada’s commitments to the international community and its adherence to the core principles of the Women, Peace, and Security (WPS) agenda[ii], Canada can help build an effective regulatory framework that is sensitive to gender-based and racialized discrimination.

Indeed, AI is not a neutral technology: its use can perpetuate numerous biases and reinforce gender- and race-based discrimination[iii] — which must be prevented before these systems are widely integrated into the military’s defence apparatus.

In 2026, AI is already being used in military decision-making, as is the case in the U.S. and Israeli militaries[iv]. Predictions made by systems such as Maven or Lavender[v]  have significantly accelerated the “kill chain” and increased the frequency of airstrikes compared to conflicts that predated the use of AI[vi]. This is drastically transforming the conduct of war and requires the regulation of the uses of AI within warfare, ensuring its compliance with the principles of international humanitarian law, and monitoring its effects on human security.

Canada now has the opportunity to engage in collaborative and transnational work on regulating the use of AI-DSS and to accordingly expand and update its national AI strategy. While these systems are mentioned in the strategy, there is no moratorium or regulatory proposals regarding their potential deployment. Finally, the non-proliferation and regulation of military technologies must be a major policy priority for the Canadian government. Canada has an important international role to play. This requires greater cross-border, multisectoral collaboration and an update to its current AI strategy.

International context

There is currently no binding instrument governing the use of AI in the military

To date, the use of AI in the military has not been regulated by states or relevant international bodies. The adoption of Resolution 78/241 “Lethal Autonomous Weapons Systems” by the United Nations General Assembly on December 22, 2023[vii], is a first step, but it is not sufficient to ensure compliance with international humanitarian law in the conduct of war and does not directly address the specific challenges associated with the use of AI in the military and AI-DSS more specifically[viii].

Obstacles to human judgment and legal liability

According to legal experts specializing in international humanitarian law[ix], military personnel and states are legally responsible for any action taken in the context of hostilities between belligerents. In this context, human judgment is essential for determining whether an action or decision complies with international humanitarian law. However, the use of AI-DSS risks eroding this judgment[x]. As demonstrated by the example of the Israeli Defense Forces’ Lavender, human verification of AI predictions is limited, if not nonexistent, and deeply biased[xi]. Furthermore, the decline of human judgment in military decision-making makes it more complex to assign legal responsibility in cases of violations of international humanitarian law[xii].

Existing violations of IHL

The recent use of AI-supported decision-making systems by the United States and Israel since 2023 has resulted in significant civilian casualties and damage to non-military infrastructure. Certain core principles of international humanitarian law—namely humanity, distinction, precaution, and proportionality—have been violated. Some systems even allow for abuses. For example, even before Israel’s Lavender was deployed in Gaza in 2023, investigations showed “that 10 percent of the human targets designated for assassination did not belong to the armed wing of Hamas at all.” [xiii] Furthermore, the lack of transparency surrounding the implementation of the algorithms used in such systems makes it impossible to verify their compliance with international humanitarian law.

Gendered and racial discriminations

Whether in the civilian or military sphere, predictions made by AI systems replicate gender- and race-based discrimination that can jeopardize civilian safety[xiv]. The data used by AI is drawn from information generated by human beings who themselves harbour biases. Consequently, “the machine can be affected by the same problems that affect humans, whether due to design flaws, deliberately biased design, or biased data resulting from human actions.” [xv]

Policy recommendations

Along with the promotion of a moratorium on the use of AI in the defence sector, Canada can develop and promote the following best practices in the international arena with the aim of establishing global standards for the use of AI-DSS. Canada may also choose to adopt this framework if it develops and implements such systems.

Work collaboratively toward a legally binding instrument within the relevant international forums

Canada has historically distinguished itself as a diplomatic and normative actor at the forefront of disarmament, arms control, non-proliferation, and the promotion of human rights. Already active in discussions on military AI and LAWS, Canada must continue its regulatory efforts regarding AI-DSS through bodies such as the Office for Disarmament Affairs (UNODA) and the United Nations Institute for Disarmament Research (UNIDIR) and highlight the expertise of Canadian and international researchers and civil society networks specializing in these issues.  Incorporating algorithmic disarmament and non-proliferation into Canada’s national action plan on the Women, Peace, and Security agenda could be a first step toward a broader political and regulatory process.

Support the monitoring of AI-DSS already in operation

Researchers, journalists, and civil society organizations are working to document the current uses of AI systems in contemporary conflicts[xvi].  The use of these systems results in a clear lack of transparency regarding how they operate, how they are used, and how they are funded. Canada can support these efforts by funding projects aimed at monitoring the use of AI systems in warfare or by developing an independent transnational body dedicated to monitoring technological advancements related to AI-supported decision-making systems.

Should Canada wish to integrate AI-DSS into its military, the following recommendations must be considered:

Require and verify compliance with international humanitarian law in the procurement and adoption of AI-DSS

According to interpretations by legal experts in international humanitarian law, states bear legal responsibility for ensuring that their military equipment complies with international humanitarian law. Consequently, in the context of its partnerships with AI companies and the procurement of military decision-support systems, a state must ensure that these companies and systems can comply with international humanitarian law. When making a purchase, the Department of National Defense and the Canadian Armed Forces must be able to verify the assurances provided by the companies.[xvii].

Establish rigorous and independent mechanisms to evaluate the collection and analysis of data by AI-DSS

Data collection and analysis are fundamental to the operation of AI systems. To guard against gender, racial, or socioeconomic biases that could lead to potential violations of international humanitarian law through the erroneous identification of people or objects[xviii], it is necessary to pay close attention to how data is collected and how it is interpreted by operators[xix]. Furthermore, the data used to train these systems can quickly become obsolete in the context of actual military operations[xx], leading to serious targeting errors. Consequently, rigorous mechanisms for evaluating the processes by which algorithms collect and analyze data must be a priority throughout the entire lifecycle of AI-DSS.

Strengthen staff training and efforts to prevent gender-, race-based and automation bias when using AI-DSS

Automation bias reinforces operators’ and soldiers’ trust in technology, particularly in the predictions generated by AI-DSS[xxi]. Similarly, operators may reintroduce certain biases into the systems. To limit the harmful effects of this “feedback loop,” the first step is to ensure that human personnel using AI-supported decision-making systems within the chain of command receive ongoing training. Furthermore, similar to the proposed guardrails associated with the acquisition of such systems, it is necessary to ensure that their initial modeling parameters allow for the reduction and prevention of potential biases[xxii].

As AI is increasingly integrated into the operations and chain of command of the armed forces, political action is needed to regulate and control it. Through its involvement in the Women, Peace, and Security agenda, Canada has already demonstrated sensitivity to gender- and racial-based discrimination in war. Given that the use of algorithms exacerbates these issues, it is essential for the government to address this challenge and make it a programmatic and political priority by 2027.

End Notes

[i] National Defense. 2024. Artifical Intelligence Strategy. Ottawa. https://www.canada.ca/content/dam/dnd-mdn/documents/reports/ai-ia/dndcaf-ai-strategy.pdf

[ii] The third national action plan is currently being implemented across several branches of the Canadian government.

[iii] Jorge Casillas. 2023. « Bias and Discrimination in Machine Decision-Making Systems ». In Ethics of Artificial Intelligence, edited by Francisco Lara et Jan Deckers. Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-48135-2_2.

[iv] Bethan McKernan and Harry Davies. 2024. « ‘The Machine Did It Coldly’: Israel Used AI to Identify 37,000 Hamas Targets ». World News. The Guardian, April 3. https://www.theguardian.com/world/2024/apr/03/israel-gaza-ai-database-hamas-airstrikes ; Lewis-Kraus, Gideon. 2026. « How Project Maven Put A.I. Into the Kill Chain ». Under Review. The New Yorker, April 15. https://www.newyorker.com/books/under-review/how-project-maven-put-ai-into-the-kill-chain.

[v] Lavender and Maven are two AI-powered military decision-support systems used by Israel and the United States, respectively. They rely on extensive surveillance networks and help dramatically speed up decision-making and the identification of military targets. See: Gideon Lewis-Kraus. 2026. « How Project Maven Put A.I. Into the Kill Chain ». The New Yorker, April 15. https://www.newyorker.com/books/under-review/how-project-maven-put-ai-into-the-kill-chain ; Rehak, Rainer, and Taylor Kate Woodcock. 2026. “Automating civilian harm: On Israel’s use of the AI-enabled targeting system Lavender in Gaza and International Humanitarian Law” Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency (New York, NY, USA) https://doi.org/10.1145/3805689.3812357.

[vi]Craig Jones and Helen M. Kinsella. 2026. « Iran War Shows How AI Speeds up Military ‘Kill Chains’ ». The Conversation, March 17. https://doi.org/10.64628/AB.5fmqfh6pp.

[vii] United Nations General Assembly. 2023. Resolution A/RES/78/241, “Lethal Autonomous Weapons Systems.” New York. December 28.https://docs.un.org/fr/a/res/78/241

[viii] Stop Killer Robots. 2026. Digital Dehumanisation and Automated Harm. New-York. https://www.stopkillerrobots.org/wp-content/uploads/2026/06/Digital-Dehumanisation-and-Automated-Harm.pdf

[ix] See Protocol I of June 10, 1977, supplementary to the Geneva Convention of August 1949.

[x] Stop Killer Robots. 2026. Digital Dehumanisation and Automated Harm. New-York. https://www.stopkillerrobots.org/wp-content/uploads/2026/06/Digital-Dehumanisation-and-Automated-Harm.pdf

[xi] Amjad Iraqi. 2024. « ‘Lavender’: The AI Machine Directing Israel’s Bombing Spree in Gaza ». +972 Magazine, avril 3. https://www.972mag.com/lavender-ai-israeli-army-gaza/ ; Rainer Rehak, and Taylor Kate Woodcock. 2026. « Automating civilian harm: On Israel’s use of the AI-enabled targeting system Lavender in Gaza and International Humanitarian Law ». Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency (New York, NY, USA), FAccT ’26, June 25, 4885‑98. https://doi.org/10.1145/3805689.3812357.

[xii] Wen Zhou et Anna Rosalie Greipl. 2024. « AI in Military Decision-Making: Supporting Humans, Not Replacing Them ». Humanitarian Law & Policy Blog, August 29. https://blogs.icrc.org/law-and-policy/2024/08/29/artificial-intelligence-in-military-decision-making-supporting-humans-not-replacing-them/.

[xiii] Amjad Iraqi. 2024. « ‘Lavender’: The AI Machine Directing Israel’s Bombing Spree in Gaza ». +972 Magazine, avril 3. https://www.972mag.com/lavender-ai-israeli-army-gaza/ ; See: Rainer Rehak et Taylor Kate Woodcock. 2026. « Automating civilian harm: On Israel’s use of the AI-enabled targeting system Lavender in Gaza and International Humanitarian Law ». Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency (New York, NY, USA), June 25, 4885‑98. https://doi.org/10.1145/3805689.3812357.

[xiv] Airwars and AINOW. 2026. Anatomy of a Kill Chain. Août. https://ai-killchain.airwars.org/ ; Rainer Rehak et Taylor Kate Woodcock. 2026. « Automating civilian harm: On Israel’s use of the AI-enabled targeting system Lavender in Gaza and International Humanitarian Law ». Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency (New York, NY, USA), June 25, 4885‑98. https://doi.org/10.1145/3805689.3812357.

[xv] Jorge Casillas. 2023. « Bias and Discrimination in Machine Decision-Making Systems ». In Ethics of Artificial Intelligence, edited by Francisco Lara and Jan Deckers. Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-48135-2_2.

[xvi] See for example : Airwars and AINOW. 2026. Anatomy of a Kill Chain. August. https://ai-killchain.airwars.org/

[xvii] Netta Goussac and Vincent Boulanin. 2026. Responsible Procurement of Military Artificial Intelligence. SIPRI. https://www.sipri.org/publications/2026/other-publications/responsible-procurement-military-artificial-intelligence.

[xviii] Laura Brunn and Marta Bo. 2025. Bias in military artificial intelligence and compliance with international humanitarian law. SIPRI. https://www.sipri.org/sites/default/files/2025-08/0825_ai_military_bias.pdf

[xix] I define an “operator” as any person employed by a military force who is responsible for verifying the predictions made by AI.

[xx] ICRC. 2024. Decisions, decisions, decisions: Computation and artificial intelligence in military decision-making. Geneva. https://www.icrc.org/en/publication/decisions-decisions-decisions-computation-and-artificial-intelligence-military-decision

[xxi] Yee-Kuang Heng. 2025. « Upskilling Human Actors against AI Automation Bias in Strategic Decision Making on the Resort to Force ». Cambridge Forum on AI: Law and Governance 1 (January): e53. https://doi.org/10.1017/cfl.2025.10029.

[xxii] Ingvild Bode et Ishmael Bhila. 2024. « The Problem of Algorithmic Bias in AI-Based Military Decision Support Systems ». Humanitarian Law & Policy Blog, September 3. https://blogs.icrc.org/law-and-policy/2024/09/03/the-problem-of-algorithmic-bias-in-ai-based-military-decision-support-systems/.