Japan weighs frontier AI for active cyber defense
Governments are starting to answer the AI-attack problem with AI defense: Tokyo says it can no longer do without frontier models, and a Huawei-and-DJI pedigree startup just raised big money to put embodied intelligence in the home.
Japan is considering folding frontier AI systems into its "active cyber defense" policy, with a senior government official pointing to attack risks from models like Anthropic's Claude Mythos. As concern grows that AI models with elite vulnerability-hunting ability could be turned into offensive cyber weapons, Japan's National Cyber Director Yoichi Iida said using frontier AI for defense is "unavoidable," according to Kyodo. The move would extend the active cyber defense framework Tokyo put in place last year — which lets the government pre-emptively disrupt malicious networks — with models that can hunt and neutralize threats faster than human teams. It fits a broader push Nikkei has tracked: the Defense Ministry plans to bake cybersecurity software into air-defense radars and fighter jets by 2027, and Tokyo is separately weighing "continuous" AI legal reforms to counter Mythos-level threats.
The signal here is bigger than one country's procurement plan: governments are concluding that only frontier AI can defend against frontier AI, and they are saying so out loud. Japan is one of the first to publicly tie its defensive posture to a named model's capability — the same dynamic that now has labs, regulators and defense ministries moving in lockstep. Whether "active cyber defense" stays defensive once frontier models are inside it is the question that will define the next year of AI governance.
Nuoyin Intelligence, the Shenzhen consumer-robot startup founded by a former head of generative AI at Huawei's Noah's Ark Lab, raised a 500 million yuan (about $70 million) angel++ round led by Matrix Partners China. Redpoint China Ventures, SenseTime Guoxiang Capital, Huadeng Technology and L2F LightSource Founders Fund joined, with LightSource Capital as sole adviser. Founded last August by Li Yinchuan — five years at Huawei Noah's Ark, where he led the generative LLM direction — alongside a co-founder who ran product on DJI's flagship lines, the company builds GLOW, a generative embodied foundation model, and KNOWIN-X1, a foldable dual-arm home robot that collapses to under 40 centimeters and debuted at WAIC. The bet is that consumer homes, not factories, are where embodied AI's long-tail generalization gets proven — and blue-chip investors are paying angel-stage prices for that thesis, with L2F following four rounds in a row. We covered the wider China robotics rally when Unitree priced its Shanghai IPO at a $9 billion valuation this weekend — Unitree prices $9B humanoid IPO, first in mainland China.
What to watch: whether Japan's active cyber defense gains an explicit AI mandate — and whether KNOWIN-X1 ships at a price a home user will actually pay.
If governments are rushing to deploy the same frontier models they fear as attack tools, does AI defense make us safer or just escalate? Tell us in the comments.
Sources: Kyodo via 金融界 (Sohu) · Sina Finance 7x24 · 金融界 AI Telegram (NetEase) · Nikkei Asia · 36Kr · 投中网 via Sina Tech · 猎云网 via Sohu · 投资界 via NetEase · AI Midday — Unitree IPO
