Information, Not Execution: Generative AI, the Two Barriers to CBRN Terrorism, and the Limits of Arms Control

Chemical, biological, radiological, and nuclear (CBRN) terrorism has been deterred by barriers running from knowledge to execution. This article argues that generative AI acts on only one: it collapses the information barrier to CBRN-relevant activity, not the execution barrier for physical attacks. Across a structured comparison of six institution-verified AI-enabled cases (2022–2025), the information barrier degrades in every case, whereas the execution barrier gives way only where execution is itself informational (the cyber domain) and remains binding for physical chemical and biological agents. We derive a four-type threat typology and apply the Multiple Streams Framework to the resulting policy window.