John Z Black

John Z Black Mar 25, 2026 AI Security #ai #democracy #japan #team-mirai #politics #schneier

Eleven seats in Japan’s House of Representatives. Nearly four million votes. That’s not a tech novelty story. That’s a proof of concept.

Team Mirai, founded by AI engineer Takahiro Anno in May 2025, demonstrated something the political world has been theorizing about for years: AI can scale genuine constituent engagement, not just ad targeting. Bruce Schneier came down on the optimistic side. His take: Team Mirai used these tools to enhance participation, not replace it. That distinction matters.

The AI Interviewer is the centerpiece. Since 2025 it’s logged roughly 8,000 hours of conversations with constituents. Not surveys. Conversations. Policy positions were built from that input, at least officially. Volunteer coordination ran on a gamified AI system during the campaign’s final stretch, hitting about 100,000 organizer actions per day. That’s not a number you reach with spreadsheets.

The optimistic case holds. The first use of this model was a democratic party that used AI to let more people have a voice, won real seats, and did it transparently enough for Schneier to call it encouraging.

But nobody audits the AI Interviewer. We don’t know what it’s optimized for, whether it nudges conversations, or how the synthesis from constituent input to party position actually works. And more importantly: the same infrastructure works for adversarial actors. An AI system capable of 8,000 hours of authentic-feeling political conversation at scale is also capable of running disinformation at scale. The tools don’t care about intent.

Team Mirai demonstrated the model works. The playbook is now public. Who adopts it next is the real story.

The full analysis of what Team Mirai’s AI actually did, Schneier’s frame, and the harder questions about where this goes next are in the complete post.