Satellites and AI as a stopgap for crumbling nuclear arms control

A generation of negotiated treaties that steadily shrank global nuclear arsenals is unraveling, leaving a verification gap that former inspectors and analysts worry will accelerate an arms race. In response, researchers at the Federation of American Scientists have sketched a fallback: convert the dense network of commercial and government sensors into a cooperative, remote verification architecture that can substitute for intrusive on-site inspections. The concept relies on satellites, aerial sensors and other remote platforms to collect signatures of missile silos, mobile launchers and plutonium production, then uses machine-learning tools to flag changes and channel outputs to human analysts. Proponents argue this model lowers the political barrier created by states’ reluctance to host foreign personnel, while preserving some capacity to detect treaty violations or disruptive force buildups. Yet the plan faces technical and diplomatic hurdles: sensor resolution, cloud cover, camouflage and deliberate deception can all defeat automated detection, and AI systems are vulnerable to adversarial data and classification errors that could produce false alarms. Implementation would therefore require agreed data standards, mutually accepted transparency measures, independent cross-checks and legal rules to govern how flagged anomalies are handled. The timing is stark—an important bilateral limit recently lapsed, even as major powers invest in new delivery systems and newly visible silo construction emerges in several countries—so the incentive for an alternative verification path is real. Still, remote monitoring cannot fully replicate the confidence gained from boots-on-the-ground inspections: sampling, physical access and the ability to inspect declared facilities historically created trust that sensors alone struggle to produce. Operationalizing this approach would involve trade-offs: broader coverage and continuous observation versus lower certainty and higher interpretive complexity. The proposal also raises strategic risks; misread signals or rushed public attribution could inflame tensions and prompt preemptive moves. Realistic next steps include pilots using agreed sensors, shared datasets for model training, and multilateral exercises to evaluate false-positive rates under adversarial conditions. In short, satellite-and-AI verification is not a panacea but it may be the most practical route to preserve at least partial arms-control verification while diplomatic channels for traditional treaties remain stalled.