Trade-Secret-Safe Framework for Military AI Sovereignty

A recent paper on arXiv (2604.20867v1) highlights that the key challenge in military AI revolves around maintaining decision sovereignty. This refers to a state's capacity to uphold control over decision-making policies, version management, fallback behaviors, audit processes, and the approval of final actions, even when analytical components are obtained from commercial sources. The document references the 2026 dispute between Anthropic and the Pentagon, Project Maven, and the latest directives from the U.S., NATO, U.K., and intelligence agencies. It suggests a framework that safeguards trade secrets while ensuring model replaceability, human oversight, and state authority, and examines how privately managed models in military operations can enable suppliers to affect operational limits.