Software Engineering Institute

This reference architecture offers guidance for designing and testing machine learning-enabled, lethal autonomous weapons systems. It introduces "conduct assurance,” which entails the integration of behavioral rules into system design and development to ensure the system behaves as expected to engender trust. This guidance covers the complete lifecycle of these rules from how they are specified, entered into the system, decomposed into machine learning-specific values, disseminated to various components, validated, and tracked throughout development. Finally, the reference architecture provides ways of promoting the properties of testability and traceability.