According to monitoring by Dongcha Beating, the Japanese government is considering using Sakana Fugu for Self-Defense Force command and control. Yomiuri Shimbun has even listed it as one of the main candidates for 'domestic AI.' Defense use also requires the exclusion of Chinese AI, and related developers must meet Japan's security screening requirements. But this 'domestic AI' is actually quite nuanced. Fugu itself is more of an AI orchestration system: a small command model is responsible for assigning tasks, while the real work is done by closed-source large models like GPT, Claude, and Gemini behind the scenes. Sakana also explicitly acknowledged today that this command model has been trained on Alibaba's Qwen. If Fugu is truly to be used for defense purposes, the Qwen layer would need to be replaced. Sakana has already retrained the command model using Google DeepMind's open-source Gemma 4 E2B, with virtually no drop in performance. The next step is to replace it with a domestic model trained from scratch in Japan. Japan is working hard to domesticate Fugu's 'command tower,' but the models providing cutting-edge capabilities still come from the United States. This system, touted as a representative of 'domestic AI,' may ultimately become: Japanese models command, American models do the work, and Chinese models are not allowed in.