Stop Killer Robots UK is at the informal exchanges on artificial intelligence in the military domain meetings in Geneva between 15 – 17 June 2026. Below are highlights from the first day
ICRC Side Event: AI in Military Decision-Making – Advancing Responsible Development and Use
An ICRC side event at a UN conference in Geneva brought together representatives from governments, industry, and humanitarian organizations to discuss the responsible development and use of artificial intelligence in military decision-making. Speakers emphasized the need to ensure that AI systems operate within existing international humanitarian law (IHL) frameworks while maintaining meaningful human responsibility and accountability.
Human Responsibility and Legal Accountability
Tabitha Bonney from the UK Ministry of Defence (MOD) stressed the importance of avoiding the anthropomorphizing of AI systems. Existing legal frameworks already establish responsibility and accountability for military actions, and states must ensure that human agency remains central to decision-making processes. While weapons reviews are an important legal requirement, they do not remove a state’s broader obligation to systematically assess and manage the risks associated with AI-enabled capabilities.
The MOD highlighted the role of procurement standards, defence requirements, and data standards in ensuring responsible deployment. Ongoing engagement with industry, including training and support throughout a system’s lifecycle, was identified as essential.
Humanitarian Concerns and System Performance
Heidi Kandiel, lawyer at the International Committee of the Red Cross (ICRC) raised concerns about how AI systems are evaluated and deployed. She noted that accuracy scores alone provide an incomplete picture and may not reflect real-world operational performance. While AI may increase speed and efficiency, these advantages can also increase risks to civilians if systems are used without sufficient understanding of their limitations.
The ICRC emphasized that humans and AI fail in different ways, many of which remain insufficiently understood. Questions such as “better at what?”, “under what conditions?”, and “with what implications for civilians?” must be addressed before AI systems are relied upon in military contexts. Concerns were also raised regarding the opacity of AI systems and the lack of transparency about how and where they are used.
Governance, Transparency and Strategic Decision-Making
The Moderator Dr. Zena Assaad noted that military decision-making takes place in inherently complex and uncertain environments. She argued that clear governance frameworks are necessary to establish expectations for the use of AI capabilities and to provide mechanisms for oversight and review. Transparency regarding governance arrangements can help build confidence among both military personnel and the public.
Speakers agreed that governance should extend beyond technical safeguards and include broader ethical and operational considerations. Effective governance frameworks should allow AI capabilities to be assessed conceptually, operationally, and legally before deployment.
Industry Perspectives
Michael Karimian of Microsoft observed that industry is already developing AI systems that can have significant real-world impacts. Because many AI technologies are dual-use, serving both civilian and military purposes, clear expectations regarding responsible development and deployment are increasingly important.
Karimian highlighted corporate responsibilities such as human rights due diligence, customer and use-case reviews, and safeguards against misuse. He argued that there remains a lack of shared understanding about what responsible corporate conduct should entail in the military AI context and called for greater clarity regarding industry obligations.
Emerging Recommendations
The discussion highlighted the importance of multi-stakeholder cooperation between states, industry, and humanitarian actors. Recommendations included:
- Designing user interfaces that encourage questioning and verification of AI outputs.
- Providing training to help operators understand when AI recommendations should be challenged or disregarded.
- Assessing whether a particular military task is appropriate for AI before introducing technological safeguards.
- Establishing feedback loops between procurement processes, legal reviews, operational doctrine, and technical governance mechanisms.
Participants stressed that responsible AI governance cannot be reduced to technical performance metrics alone. Broader legal, ethical, and humanitarian considerations must remain central to international discussions.
Key Takeaway
A recurring theme throughout the event was the importance of assessing whether AI is genuinely fit for purpose in military applications. As one ICRC representative noted, AI should not be viewed as a solution to underlying challenges; rather, it acts as a “magnifying glass,” amplifying existing strengths, weaknesses, and risks. Speakers agreed that the central question is not simply what AI can do, but whether a particular task should be delegated to AI in the first place, particularly where civilian protection and compliance with international humanitarian law are concerned.