September 21, 2026
Stephen Herzog
The following is an excerpt from The Nonproliferation Review.
Artificial intelligence (AI) is now a central element in many of the most pressing debates about nuclear, biological, and chemical weapons. Yet, the scholarship on AI and weapons of mass destruction (WMD) remains highly fragmented. Much of the existing literature asks whether AI may lower knowledge barriers to dangerous activities by increasing access to expertise or accelerating sensitive research. Another strand focuses specifically on nuclear weapons and the concerns that AI integration into command-and-control architecture could jeopardize strategic stability by compressing decision time or eroding meaningful human control. A third body of work turns to monitoring and verification, where AI may offer governments and open-source intelligence analysts new avenues to assess compliance and noncompliance with arms-control, nonproliferation, and disarmament obligations. Finally, a fourth set of studies zooms in on governance institutions and whether they can keep pace with private-sector technology before AI-enabled WMD risks emerge. This introduction argues that these conversations should not be siloed. It advances a process-based approach that tracks how AI may alter the pursuit, operationalization, detection, and eventual constraint of WMD capabilities. Developments at one stage of the life cycle can also affect what happens elsewhere. This overview illustrates just how rapidly AI is reshaping the entire spectrum of WMD politics.
Continue reading at The Nonproliferation Review.
