In 2024, discussions on governing military artificial intelligence intensified, propelled by summits such as Responsible AI in the Military Domain (REAIM) and policy documents like the 2023 REAIM Political Declaration. These initiatives sought to establish norms for autonomy in weapon systems and AI-assisted targeting. Yet, the same year saw expanded deployment of AI decision-support tools in active conflicts, revealing a stark disconnect between aspirational governance frameworks and battlefield realities. Claims that AI would protect civilians or shorten wars have not been borne out; violence has escalated, civilian casualties—especially among children—have risen, and geopolitical tensions have deepened.
Professor Elke Schwarz of Queen Mary University London identifies three structural challenges undermining responsible governance: the impermanence of AI systems, the influence of private-sector financial interests, and the expansionist logic embedded in AI technology itself.
Modern AI, built on statistical data processing and machine learning, is inherently iterative. Systems require frequent updates to remain effective, particularly in adversarial environments. For example, unmanned aerial vehicles may need updates every six to twelve weeks; complex AI targeting systems likely demand even more frequent revisions. Each update risks introducing new vulnerabilities, necessitating rigorous testing and evaluation before deployment—a process at odds with wartime imperatives for speed. The rapid evolution of AI also ensures that new problems emerge with each generation of technology. Large Language Models illustrate this, bringing issues like “hallucinations” and anthropomorphizing alongside longstanding biases. Regulatory approaches that react to each new capability risk shifting focus from responsibility to mere risk management.
The second challenge lies in the market dynamics shaping military AI. Since 2021, venture capital has injected approximately USD 130 billion into military technology startups, many from outside traditional defense circles. VC investment operates on compressed timelines and thrives on the fail-and-iterate ethos common in Silicon Valley, favoring rapid deployment over cautious evaluation. This culture prizes disruptive innovation and market capture, sometimes at odds with established norms designed to restrain the use of force. Lobbying efforts and recruitment of former military and policy officials help align regulatory attitudes with commercial goals. As Schwarz notes, “Unless we acknowledge the tension in interests by the various invested stakeholders in this military AI domain, effective governance is, to put it bluntly, unlikely.”
The third factor is the expansionist nature of AI itself. Technologically, AI systems perform best when integrated with other AI systems and fed vast amounts of relevant data. Philosophically, this mirrors Günther Anders’ 1988 observation: “Every machine is expansionistic, that is to say, imperialistic; each creates its own service – and colonial empire.” In practice, once embedded, AI systems tend to drive mission creep. Decision-support tools may evolve from recognizing known targets to discovering new ones, flagging “suspicious” actors across global theaters. This recalls the “disposition matrix” approach from the early 2010s, where predictive targeting extended far beyond traditional battlefields. Such expansion risks amplifying suspicion and hostility rather than reducing conflict.
The United Nations Secretary-General Antonio Guterres has warned that “human agency must be preserved at all cost” in the face of autonomous weapons. This imperative applies equally to AI-assisted targeting, where human judgment can be eroded by system logic and scale. Multi-domain, interconnected AI tools promise efficiency but may also normalize constant surveillance and preemptive action, deepening mistrust among nations.
While multi-stakeholder forums remain vital, their success depends on a willingness to refrain from deploying AI when risks outweigh benefits, to confront financial interests that conflict with humanitarian aims, and to resist the technological momentum that drives ever-wider application. Without such restraint, the notion of “responsible AI in the military domain” risks becoming an illusion, overshadowed by the incommensurability between AI’s operational realities and the ideals of ethical warfare.