Summary
- Most political use of AI mirrors old broadcast advertising — one message pushed to many — rather than genuine constituent engagement.
- Japanese party Team Mirai deployed an AI interviewer that generated over 300,000 messages across 16,000 conversations, with policy changes cited directly in legislative hearings.
- Scottish government-funded CrownShy and several US academic and non-profit platforms offer open-source or commercial tools for structured, AI-facilitated deliberation at scale.
- These ‘broad listening’ tools are designed to surface community input rather than amplify partisan messaging, and several are open-source.
- A Democratic-aligned campaign investment firm has signalled interest in funding AI-native systems focused on authentic, bottom-up community insights.
The Broadcast Problem
Security professionals spend considerable time thinking about how technology gets misused. The AI-in-politics story is a useful mirror. As Bruce Schneier and Nathan E. Sanders argued in The Guardian, the dominant use of AI in US political campaigning is not substantively different from television advertising or web banner ads: a candidate’s message is pushed outward to as many people as possible. The technology has changed; the underlying model has not. That model treats voters as targets rather than participants.
What ‘Broad Listening’ Actually Means
The more interesting application — and the one with documented results — is what Schneier and Sanders call ‘broad listening’. These are AI tools designed to collect public input in richer formats than a checkbox survey, conduct extended conversations with constituents, synthesise the results, and feed that synthesis back into policy decisions. The key design principle is that the system is oriented toward intake, not output.
Team Mirai: A Working Example
Japan’s Team Mirai, a newly founded political party, built an AI interviewer to gather constituent views on specific legislation. Voters engage in extended conversations with the chatbot about policy issues, and the party has scaled this across a wide array of bills through an AI-powered portal. The numbers are not trivial: more than 300,000 messages exchanged across 16,000 AI interviews. Critically, the party has demonstrated that the input is actually used — members have cited AI interview findings during legislative committee hearings, published syntheses of voter input, and amended policy platforms based on what they heard. Team Mirai has since won 12 seats in Japan’s Diet. The party describes itself as a ‘utility party’, building tools intended for any political party to use, not just itself.
Scotland, Stanford and MIT
CrownShy, funded in part by the Scottish government, is developing a platform called Comhairle that combines AI interviewing with tools for synthesising viewpoints, running virtual assemblies, and sharing video testimonials. Like Team Mirai’s stack, it is open-source. In the United States, several institutions have piloted comparable tools: Stanford-affiliated deliberation.io has been used for public listening sessions by Washington DC; the MIT-affiliated Cortico project surfaces under-heard community perspectives from recorded conversations and is running listening sessions at libraries; and the non-profit-built Talk to the City applies AI to large datasets of stakeholder input. Commercial operator Remesh has been tested in policy development scenarios as well.
Open Source and Partisan Funding: An Interesting Tension
A notable feature of both Team Mirai’s tools and CrownShy is that they are open-source despite being funded by political parties. The authors describe them as built to improve democratic processes broadly, rather than to deliver partisan advantage. That is a meaningful design choice, and one that is verifiable — in principle — because the code is available for inspection. Whether that holds in practice over time is a different question, and not one the available sources resolve.
The US Signal: Investment
One indicator that this approach may gain traction in US politics came from Higher Ground Labs, a Democratic-aligned campaign technology investment firm, which launched a fund targeting ‘AI-Native Campaign Systems’ and ‘community-led messaging platforms that surface authentic, bottom-up insights from real conversations’. Whether investment translates into deployed tools at meaningful scale remains to be seen.
The Broader AI Policy Debate
Schneier and Sanders are careful to separate AI as a technology from AI’s commercial interests. They advocate for accountability measures against AI companies for harms caused by their models, taxation of revenues, and even potential nationalisation if the AI market bubble bursts. At the same time, they argue that reflexive rejection of the technology by sceptics, and uncritical embrace by boosters, both miss the point. The tools themselves are not inherently good or bad; the design intent and governance structures around them are what matter.
Why it matters
For CISOs, the relevance here is layered. First, any organisation — government body, regulated industry, large enterprise — considering AI-facilitated stakeholder engagement needs to understand the data governance and privacy implications of running extended AI conversations with constituents or customers at scale. These systems collect sensitive input; how it is stored, who can access it, and how synthesis models handle it are legitimate risk questions. Second, the open-source nature of tools like Comhairle and Team Mirai’s stack means security teams may encounter them in public sector or civic technology contexts and should be prepared to assess them. Third, the AI-in-disinformation risk that sits alongside this story — deepfakes, AI-generated propaganda — remains a live threat to any organisation with a public profile or stakeholder communications function.
What to do now
- If evaluating AI-facilitated stakeholder engagement tools, assess data governance frameworks for how constituent or customer conversation data is stored, retained, and used to train or refine models.
- Review open-source civic AI tools (such as those from Team Mirai or CrownShy) against your organisation’s software supply chain and third-party risk standards before any adoption in public sector contexts.
- Distinguish between AI tools designed for broadcast (outbound messaging) and those designed for broad listening (structured intake) when assessing vendor claims — the risk profiles and data flows differ substantially.
- Monitor the AI-in-disinformation space alongside constructive AI use cases, as both are active in the same political and public communications environment your organisation operates in.
