A global AI safety regime should learn from nuclear arms control by focusing on the physical infrastructure that enables strategic capabilities. Instead of just seeking promises, it should aim to control access to chokepoints like advanced chip manufacturing and the massive data centers required for frontier models.

While the US military opposes bans on autonomous 'killer robots' for conventional warfare, it maintains a firm 'human-in-the-loop' policy for nuclear launch decisions. This reveals a strategic calculation: the normative value of preventing autonomous nuclear use outweighs any marginal benefit, a line not drawn for conventional systems.

The military is applying powerful AI software for intelligence and targeting, but the physical hardware—planes, missiles, and interceptors—was not designed for this new reality. This mismatch creates inefficiencies, such as using expensive Patriot missiles designed for jets to shoot down cheap drones, highlighting a hardware-software gap.

The debate around AI in warfare often misses that significant autonomy already exists. Systems like the Phalanx Gatling gun and "fire-and-forget" missiles, which operate without human supervision after launch, have been standard for decades, representing a baseline of existing automation.

Developing nuclear weapons is technically difficult. AI can lower this barrier by optimizing complex processes like centrifuge design, explosives modeling, and supply chain management. It can also help nascent programs evade export controls, making a bomb more attainable for smaller states without established nuclear industries.

The popular comparison of AI to nuclear weapons has a critical flaw. Nuclear regulation relies on tracking scarce, physical, and interceptable fissionable materials. AI, as software and weights, can be copied and distributed far more easily, making the nuclear non-proliferation playbook a poor and dangerous model for AI governance.

The immense strategic advantage offered by AI ensures its development will continue, regardless of safety concerns from insiders. Much like the Manhattan Project, which proceeded despite catastrophic risk, the logic of "if we don't, China will" makes unilateral cessation of research impossible for any major power.

The debate over autonomous weapons is often misdirected. Humanity has used autonomous weapons like landmines for centuries. The paradigm shift and true danger come from adding scalable, learning "intelligence" to these systems, not from the autonomy itself.

AI targeting systems excel at generating vast target lists for rapid, shock-and-awe campaigns. However, they are currently being applied to a slower, attritional conflict. This misapplication turns operational excellence into a strategic dead end, where the machine simply produces more targets without a causal link to defeating the enemy.