Satellite remote sensing is increasingly recognized as a critical tool for arms control and nonproliferation missions. Recently, there has been growing interest in leveraging artificial intelligence (AI) and machine learning to enhance analytical efficiency. However, this integration introduces significant risks due to the susceptibility of AI models to counter-AI attacks that may compromise their accuracy, reliability, and security. This article develops a typology of emerging threats posed by counter-AI threats against AI-driven satellite remote sensing, categorizing them in four primary scenarios: data poisoning, model evasion, data inference, and model extraction. These threat vectors span a spectrum of techniques, both digital and physical, raising vital security concerns. To address these risks, the article proposes a comprehensive defense framework centered on five pillars: controlling access and quality of data and models, enhancing the robustness of AI frameworks, advancing system monitoring capabilities, fostering knowledge sharing and threat awareness, and integrating adaptive and resilient risk management. The core message: AI functions not only as a force multiplier but also as a threat multiplier, transforming AI-enabled arms control, nonproliferation, and peacekeeping missions into increasingly dynamic hider-seeker games.
Without the quantitative constraints of New START or any other formal nuclear arms control framework, states must navigate future conflicts and crises with less clarity and confidence. This is an unwelcome development for strategic stability at an inopportune time globally. But even if the expansive US and Russian arsenals had remained capped, the greatest risks incurred over the last decade and surely into the next are qualitative. These risks owe much more to the understandings between leaders on acceptable roles of nuclear weapons, the specific capabilities fielded in national arsenals, and the broader risk profiles of day-to-day operations than weapon counts. There is little historical evidence that rapid surges in nuclear weapons technologies or deployments yield strategic stability; rather, there is a consistent record that unconstrained capability additions have led to serious instabilities, requiring new arms control efforts to stabilize the deterrence relationship. Putting aside any near-term reconstitution of formal arms-control mechanisms, states can better navigate the next decade by focusing on the qualitative elements of nuclear strategy. By unilaterally constraining weapon types, avoiding entanglement risks, and upholding norms of restraint, states can construct a more stable base from which to navigate the myriad emerging security threats of the late 2020s and beyond.
In the aftermath of New START, great-power arms control is at a crossroads. Many observers would like to see a new system established, possibly including China. However, an examination of historical arms control agreements strongly suggests that conditions do not currently favor the establishment of a new framework, because acute competition in nuclear armament is not currently underway. On the other hand, there is an urgent need for risk-reduction measures to address emerging technological and geopolitical trends which threaten to destabilize nuclear deterrence relationships. Governments, academics, and advocates should focus on this area in the near future.
Arms control is in crisis, but there is enough empirical evidence to suggest that a world without arms limitations is a much more dangerous place. Nuclear weapons states have a special responsibility to reverse the current trends.
