When the World AI Cooperation Organization launched in Shanghai on July 16, with 29 signatory nations and the UN Secretary-General in the room, the coverage mostly focused on the optics. Fair enough, the optics matter. But I think the more interesting story is the problem WAICO was created to sit on top of, because that problem isn’t going away no matter how many organizations we build around it.
Here it is in plain terms: every state has the legal right to modernize its military with whatever technology exists, and AI is now that technology. Nobody is going to sign that right away. At the same time, nobody —including the states doing the modernizing — actually wants an unvetted algorithmic arms race. Those two facts have to coexist somehow, and pretending otherwise is why so much AI governance talk goes nowhere.
I’ll say the unpopular part directly: blanket bans on military AI are unenforceable, and everyone negotiating them knows it. No country trades away its security for an ethics document. If WAICO and the frameworks around it want to matter, they need to skip the aspirational language and lock in a small number of structural guardrails that states can realistically accept. I’d argue there are three.
Keep a Human in the Kill Chain. There’s a real difference between using AI for predictive logistics or radar processing, which is just modernization, and fielding weapons that select and engage human targets on their own. Lethal autonomous weapons systems put the decision to take a life inside an algorithm, and once you’re there, accountability has nowhere to land. No commander, no operator, no state official you can point to. Meaningful human control isn’t a constraint on defense. It’s the thing that keeps defense legally and morally coherent, and it should be treated as a hard red line, not a talking point.
Stay ahead of the geopolitical week.
MD Briefing delivers expert analysis across five global fronts — the Indo-Pacific, energy, geoeconomics, European security, and the Middle East — every Monday morning. Free.
Build Crisis Protocols Before the Algorithms Meet. Military AI operates at machine speed. In a tense theater, two competing systems reacting to each other’s outputs, possibly on bad data could escalate in milliseconds, long before any human sees what’s happening. We spent the Cold War building hotlines for exactly this reason, just slower. The equivalent now is shared technical crisis-communication channels that connect Western frameworks with newer platforms like WAICO. Unglamorous work, but it’s the difference between an incident and a war started by a miscalculation nobody ordered.
Treat the Global South seriously as rule-makers. One of the notable things about WAICO’s founding membership is how many developing nations showed up wanting to write the rules rather than inherit them. That instinct is correct. In a fully unregulated military AI landscape, countries without hyperscale infrastructure don’t just fall behind, they become soft targets for cyber operations and AI-driven espionage, and collateral damage in conflicts they’re not party to. Governance that only protects the states that already have the compute isn’t governance. Defensive resilience and capacity-building have to be part of the package.
There’s one more piece, and it’s physical rather than political. Data centers and critical digital infrastructure are the foundations everything else runs on, and right now their status in wartime is dangerously ambiguous. We should be pushing for international consensus that treats them as protected non-military assets, the same way we treat hospitals and dams. Without that, every guardrail above sits on sand.
WAICO’s launch confirms that the architecture of AI governance is being written right now, in real time, and not only in Washington or Brussels. The goal of these platforms shouldn’t be to limit any nation’s right to defend itself, that fight is unwinnable anyway. The goal is narrower and more urgent: make sure that in securing our borders, we don’t build a system of autonomous chaos that no one can switch off. Sovereignty stays. But it needs human oversight, technical transparency, and a shared willingness to restrain the machines before the machines remove the option.
