A new position paper argues that artificial intelligence researchers must step into a leadership role in shaping international arms control agreements for military AI systems, similar to how nuclear diplomacy has prevented catastrophic conflicts. The argument comes as defense contractors and military establishments worldwide accelerate their integration of AI into weapons systems and military decision-making, creating urgent risks that traditional security frameworks are unprepared to address.

Why Should AI Researchers Care About Military Applications?

The rapid advancement of AI capabilities has drawn significant attention from governments and militaries globally, but the focus of most AI safety research remains on long-term risks from superintelligent systems. This approach may miss the immediate dangers posed by frontier AI models, which are advanced systems at the cutting edge of capability, being integrated into defense applications right now. Military AI systems present unique challenges because of their speed, autonomy, and potential unpredictability, which create novel escalation pathways that existing deterrence frameworks may not adequately address.

The concern is not hypothetical. Armament manufacturers and defense contractors are increasingly investing in AI capabilities and forging partnerships with AI companies, creating what researchers describe as a "burgeoning coalition" that demands urgent attention. Without comprehensive ethical frameworks and international coordination specifically designed for military AI technologies, strategic instabilities between competing nations will likely intensify, creating risks that existing deterrence doctrines are fundamentally ill-equipped to handle.

What Can Nuclear Arms Control Teach Us About AI?

Arms control agreements have successfully reduced catastrophic risks throughout history by using diplomatic agreements to ensure mutually beneficial outcomes between adversaries. In the nuclear domain, these agreements have curtailed anti-ballistic missile systems, reduced warhead counts on delivery vehicles, and eliminated entire classes of nuclear weapons. These measures demonstrably decreased the likelihood and potential severity of armed conflict by removing destabilizing factors that might otherwise trigger security dilemmas or arms races.

The decades of experience with chemical, biological, radiological, and nuclear weapons control offers valuable precedents for AI governance. These historical lessons provide frameworks for verification mechanisms, confidence-building measures, and ways to distinguish between permissible and prohibited applications of dual-use technologies, which are tools or systems that can serve both civilian and military purposes. However, current arms control diplomacy and technical verification mechanisms are not prepared for the novel challenges that frontier AI models will introduce into modern warfare.

How to Bridge AI Research and Arms Control Policy

• Establish Collaborative Research: Create a new foundation of collaborative research between AI researchers and arms control diplomacy experts to develop frameworks that address military AI risks before they destabilize global security.
• Define Military AI Systems Clearly: Establish precise definitions of military AI systems, which are those developed, operated, or utilized by nation-state military institutions like the United States Armed Forces or China's People's Liberation Army, to enable targeted diplomatic agreements.
• Identify Addressable Risk Vectors: Focus arms control mechanisms on challenges that diplomacy can actually solve, including escalation risks, alignment faking where AI systems appear safe but behave dangerously, and gradual disempowerment of human decision-making in military contexts.

The researchers propose that AI safety frameworks, while essential, remain insufficient to guarantee that frontier models deployed in military contexts can reliably mitigate risks while accounting for strategic interactions between nations. The pace of anticipatory research addressing emerging challenges critically lags behind the rapid development and military adoption of advanced AI capabilities. Addressing potentially destabilizing military applications requires multinational dialogue to identify specific risks, establish constraints on certain research trajectories, and develop diplomatic agreements among potential adversaries.

One prominent example of the kind of deterrence regime that might emerge is "Mutual Assured AI Malfunction," or MAIM, a concept resembling nuclear mutual assured destruction where any state's aggressive bid for unilateral AI dominance is met with preventive sabotage by rivals. This illustrates how arms control thinking is already beginning to adapt to the AI era, though much work remains.

What Are the Barriers to Implementing AI Arms Control?

The current landscape presents a fundamental challenge: reducing global risk in military AI will introduce a new class of problems that have not been completely understood or formally defined. Arms control, in its current state, is unprepared to solve the challenges that frontier AI models will introduce into modern warfare. This gap between the speed of AI development and the maturity of diplomatic frameworks creates a critical window in which AI researchers must take a leading role in advancing arms control research.

The argument being made is straightforward but consequential: AI researchers possess the technical expertise necessary to help diplomats and arms control experts understand the specific risks posed by military AI systems. Without this collaboration, humanity will lack the capability to achieve safe military AI governance. The researchers argue that this represents a new responsibility for the AI community, one that extends beyond traditional safety research into the realm of international security and diplomatic innovation.