For the past fifty years, major nuclear powers have depended on a complex web of treaties designed to gradually reduce the number of nuclear weapons globally. With the recent expiration of the last significant arms treaty between the United States and Russia, experts are contemplating a new approach: employing satellite technology and artificial intelligence (AI) to monitor nuclear arsenals.
According to Matt Korda, an associate director at the Federation of American Scientists, this proposal is seen as a "plan B" following the dissolution of established treaties. In the report titled Inspections Without Inspectors, Korda and his coauthor Igor Morić suggest leveraging existing satellite and remote sensing technology for nuclear monitoring—functions previously handled by on-site inspectors.
Korda notes that AI could enhance this monitoring effort, especially in pattern recognition. If a substantial and well-curated dataset were available, AI could potentially identify both minor changes in specific locations and specific weapon systems.
The expiration of the New START treaty on February 5 marked a critical juncture in arms control, with the U.S. and Russia reportedly committed to maintaining the status quo for the time being. However, both nations are heavily investing in developing advanced nuclear weapons, alongside other countries such as China, which is in the process of constructing new missile silos. As trust among nations declines, the urgency for reliable nuclear monitoring grows.
Korda and Morić’s vision involves a framework where nations cooperate to utilize satellites for monitoring intercontinental ballistic missile (ICBM) silos and other nuclear facilities, avoiding the need for invasive on-the-ground inspections. This system would rely on AI to analyze the monitoring data, which would then be subjected to human review.
However, inherent challenges exist with this model. The training of AI systems necessitates robust datasets while substantial data on nuclear arsenals is sparse. There’s also the complexity of establishing what the AI should monitor—whether it’s tracking the presence or absence of particular objects, classifying what is observed, or monitoring changes over time.
As countries potentially shift toward a monitoring system reliant on AI and remote technology, they would need to collaborate on how these systems operate and the specific criteria for monitoring, which would demand new negotiations amid ongoing tensions.
Al-Sayed from the Union of Concerned Scientists highlights the complexities involved, raising concerns about the reliability of AI in monitoring nuclear arsenals. Current AI systems are still prone to errors and security vulnerabilities, which raises questions about their integrity in a high-stakes context like nuclear arms control.
Despite these limitations, Korda asserts that utilizing satellites along with imperfect AI to oversee nuclear weapons could be a vital step in preventing an arms race—a “small bridge” to a safer nuclear future. While the absence of a robust treaty regime like New START may not lead to disarmament, it could help avoid the deployment of numerous additional nuclear weapons globally.
