Researchers from various institutions have published a paper on testing and evaluation of agentic AI systems in military command and control. They analyzed 240 documented practices across eight dimensions and three lifecycle stages, identifying eight assumptions that established methods make about their test article. These assumptions are weakened by the agentic properties of the AI systems, which can lead to unreliable inference from tested to fielded behavior. The authors derive ten assurance claims for four assumption clusters and assess current and emerging methods' ability to address them through five C2 scenarios. --- Why it matters: This matters because it highlights the challenges in testing and evaluating complex AI systems that are designed to make decisions autonomously, which is crucial for military command and control applications. The findings have implications for the development of assurance cases that can support the procurement and deployment of such systems. Source: https://arxiv.org/abs/2608.20597