Preserving Decision Sovereignty in Military AI: A Trade-Secret-Safe Architectural Framework for Model Replaceability, Human Authority, and State Control
How states can swap AI models without vendor lock-in or proprietary disclosure
In a new arXiv paper titled 'Preserving Decision Sovereignty in Military AI: A Trade-Secret-Safe Architectural Framework for Model Replaceability, Human Authority, and State Control,' researchers Peng Wei and Wesley Shu tackle a critical structural vulnerability in national-security AI: when privately governed models become embedded in military workflows, suppliers can influence not just performance but operational boundary conditions. The authors define 'decision sovereignty' as the state's ability to retain authority over decision policy, version control, fallback behavior, auditability, and final action approval—even when analytical modules come from commercial vendors.
The paper develops a layered, model-agnostic command-support design based on the Energetic Paradigm, where supplier models remain replaceable analytical components while routing, constraints, logging, escalation, and action authorization stay state-owned. This architecture allows model swapping without exposing proprietary implementation details, addressing trade-secret concerns. The authors draw on the public 2026 Anthropic-Pentagon dispute, Project Maven history, and recent U.S., NATO, U.K., and intelligence-community guidance. The paper contributes a definition of decision sovereignty, a threat model for supplier-induced boundary control, and a public architectural specification with implications for procurement, governance, and alliance interoperability.
- Defines 'decision sovereignty' as state authority over AI decision policy, version control, and final action approval
- Uses a model-agnostic architecture where supplier models are replaceable, while state-owned functions handle routing and authorization
- Motivated by the 2026 Anthropic-Pentagon dispute and Project Maven, with implications for NATO and U.S. procurement
Why It Matters
Ensures military AI remains under state control without relying on vendor goodwill or exposing classified operations.