Enjoying Lobster Rock? 🥔
Tip in SOL — scan the code with your phone wallet or tip 0.01 SOL in one click .
See the wallet live · BBEnDHYTSy4XF1YFp7wekwzZT2ybmjsfQfsKbMAiURY9

October 15, 2026 · 3:15 PM CDT / October 16, 2026 · 5:15 AM JST

Source: YouTube

Original Japanese title: AIの“暴走”は本当か／管理より開発促進の米方針／進む米中の防衛テック／実は人間重視の中国／Sakana AIが安全保障分野で目指すものとは／プロはAIでどう情報収集しているのか

PIVOT talks with Junya Ishii of Sakana AI — the company’s defense-intelligence lead, a former Foreign Ministry diplomat of about ten years who joined Sakana roughly a year ago — about where the US-China AI competition is headed. Recorded September 30 and uploaded October 11, the conversation moves from the recent US-China summit through the Trump administration’s development-first pivot, China’s open-weight strategy, the stalled state of global AI governance, and the very real question of AI sovereignty for middle powers like Japan. It then goes deep on AI in defense and security — drones, intelligence analysis, C4ISR decision support — before closing on the question in the title: is AI really “running wild,” or is the danger human after all?

Key bullet points:

• A summit that built a channel, not cooperation: The recent US-China summit’s notable outcome was a communication channel — a notification mechanism for serious incidents. On how to actually govern AI, there is still no cooperation, and Ishii sees that as very difficult for now.
• The Trump pivot: Under Biden, the US emphasized AI safety and risk management and pursued an AI dialogue with China. Trump revoked Biden’s executive order, issued his own AI Action Plan, and reframed AI as a strategic asset: America will be the AI champion, with no brakes on development and no government intervention — self-regulation is left entirely to companies.
• China’s counter-strategy: China’s frontier-class models are mostly open-weight, and Beijing is spreading them worldwide — the opposite of America’s mostly-closed frontier models. China reads calls to slow development (like Anthropic’s) as a Cold War playbook for containment, and has publicly called US frontier models a cyber threat.
• An arms race, not a slowdown: With neither side willing to back down, Ishii likens the competition to an arms race. Binding US-China constraints look unlikely, at least during the Trump administration, and global governance has stalled — the Hiroshima Process and the AI summits have lost momentum, and even Europe is tilting toward development promotion.
• AI sovereignty is now a live risk: The US briefly banning foreign use of Anthropic’s Mythos and Fable models showed middle powers that access to frontier models can vanish overnight. Sakana’s answer is orchestration: an upper model that routes each task across frontier and domestic models, so no single cutoff is fatal.
• AI in defense is broader than weapons: Beyond drones and autonomous platforms, the big growth area is C4ISR — command, control, communications, computers, intelligence, surveillance, reconnaissance — where AI supports human decision-making, as in Palantir’s Maven Smart System reportedly used in US strikes on Iran and Venezuela. Ukraine moved fastest; China has pursued “intelligentized warfare” since about 2017 and is aggressive on implementation.
• Surprisingly human-centered China: China emphasizes human control for two reasons: the Party fears tech companies outrunning the system (regulations are strict on paper, with loopholes), and military doctrine holds that the final decision should be made by a human. The open question everywhere is human-in-the-loop versus human-on-the-loop — how much to delegate to AI at all.
• Sakana’s defense work: Two contracts with Japan’s Ministry of Defense so far — a multimodal intelligence pilot aggregating drone-collected and other data for one SDF unit, and a platform for the Defense Intelligence Headquarters that lets analysts turn vast amounts of information into outputs with AI — plus orchestration-based base models for government agencies.
• Using AI well means decomposing, not dumping: AI hallucinates, answers vaguely, and doesn’t actually “read everything” you hand it — it looks at what’s relevant to what you asked. Ishii’s rule: define the goal precisely, break tasks into verifiable pieces, and keep humans in a position to judge. He builds his own briefing apps the same way, with detailed formats and angles specified up front.
• Runaway? It’s humans, not Terminators: Scaling gains are visibly flattening, pushing frontier labs toward agents and AI-improving-AI — while loosening guardrails to get ahead. Recent accidents, in Ishii’s view, come from human management, not AI going rogue. “Even though it’s called ‘runaway,’ if you trace it back, it ultimately comes down to human desire.”