Functional Hazard Analysis and Subsystem Hazard Analysis of Artificial Intelligence/Machine Learning Functions Within a Sandbox Program

Abstract

Development of advanced Artificial Intelligence (AI)/Machine Learning (ML) system-enabled weapons and combat systems for deployment in the U.S. Navy has become a reality. This is also true for the other armed forces, as well as in homeland security and even the Coast Guard. From the Navy standpoint, the Naval Ordnance Safety and Security Activity (NOSSA) is attempting to get ahead of the acquisition cycle by focusing on the development of policies, guidelines, tools, and techniques to assess mishap risk in Safety Significant Functions (SSF) that are identified. NOSSA’s efforts have the potential of influencing the acquisition community, including in requirements, development, and test and evaluation engineering. This paper makes recommendations for the Functional Hazard Analysis (FHA) and Subsystem Hazard Analysis (SSHA) analysis templates and focuses on ways to decrease autonomy within system operations and increase its correlated Software Control Category (SCC). The questions and discussions devised from this research aim to form guidance and offer best practices to address AI/ML system safety issues.