How Many People Does It Take to Challenge a Government?
In July 2025, a new Japanese political party called Team Mirai published its proposals and asked voters what they would change. The party gathered feedback at speeches and meetings, and it gave people a way to talk through the proposals with an AI system. It says it used what it heard to revise its platform. Its founder, Takahiro Anno, won an upper-house seat that month, and by September 2026 Team Mirai had 11 members in Japan’s lower house. How much the AI contributed to those results is still an open question. [1,2,3]
I’ve been thinking about some version of this since GPT-2, and the part of this campaign I keep coming back to is what it took to listen to people. Somebody had to read what voters sent in, ask a follow-up if something was unclear and work out which suggestions belonged in the platform. How much of that could the party have done with the same people and budget if every exchange needed someone on staff? I want to know whether that difference is big enough for people who used to struggle to get a hearing to get a government to change course.
From getting heard to getting people to act
In 1776, printers carried Thomas Paine’s argument for American independence well beyond the people he could have reached in person. When Franklin Roosevelt went on the radio in 1933 to explain why the banks had closed and what to expect when they reopened, he could make his case directly to people across the country. A new way to reach people could help someone challenge a government, and it could give the people in office a better way to explain what they were doing. [4,5]
The internet gave people a way to write back, and it gave people who wanted to make an issue look bigger than it was a way to do that, too. In the FCC’s 2017 net neutrality proceeding, a college student used software to submit more than seven million comments under made-up identities, according to New York’s attorney general. One person could already look like a crowd. [6]
With Team Mirai, the work I’m interested in begins when a voter has a question about a proposal. A small staff can publish the party’s position once, but answering follow-ups and making sense of what people send back takes time. AI can help with both, including picking out a concern that keeps coming up even when people describe it in different ways. What I’d measure is how many conversations the party was able to have that it would otherwise have missed, and what it did with what it heard. [1]
An October 2026 study gives us a way to see whether one of those conversations can get someone to act. The researchers asked paid participants in the UK to talk to a chatbot and gave them a chance to sign a real petition. In one experiment, about 40% signed after a chatbot made the case for it, compared with 21% after a chatbot conversation about a neutral topic. For a small group running a campaign, the question is what it would cost to get people to have those conversations in the first place, and whether enough of them would act for officials to take notice. [7]
I’d welcome a group using AI to hear from people it used to miss, provided those people know who they’re dealing with. Just before New Hampshire’s 2024 presidential primary, a political consultant arranged robocalls that used an AI imitation of Joe Biden’s voice to discourage people from voting. The FCC imposed a penalty for the robocall campaign, although we still don’t know how the calls affected voting. [8]
A group can also put pressure on a government by getting into a system the public depends on. In January 2026, an attacker used commercial AI during a break-in at the water utility serving Monterrey, Mexico. Investigators found that the attacker used the software to help write and adapt tools after gaining access to the utility’s office computers. The attempt to get to the equipment controls failed, and the investigators couldn’t establish how many people were behind the intrusion. [9]
The two bets a country makes
The first is about how much a government needs from its people. If a department can get through more work with the people it has, it might use the time to catch up or hire fewer people when positions open. If it can do without work it once relied on a group of people to do, those people may have less leverage over pay. At the same time, citizens who find it easier to organize may be able to get a hearing on issues that used to go nowhere. I’d want to know how both changes play out before saying whether the government or its citizens gained more power.
The government also needs to pay for the work. Britain’s Office for Budget Responsibility looked at what would happen if more income went to profits and less went to wages. Under its assumptions, the economy could be just as big and the government could still collect less in taxes. So if a government is counting on AI to let it do more, one question is whether it can tax the companies that earn the profits. If some of the money that used to go to wages goes to a supplier abroad, the government may have less income it can tax. [10]
The second bet is about what a government can control inside its borders. On June 12, 2026, Anthropic said the U.S. government had directed it to cut off foreign nationals’ access to Fable 5 and Mythos 5, including for people inside the United States. Anthropic said it couldn’t verify nationality in real time, so it suspended both models for everyone. Fable was available again globally through Anthropic on July 1. In the meantime, the order had reached users all over the world because it went through the U.S. company supplying the models. [11,12]
In The Invisible Interface, I write about what happens when we give software the records and permissions it needs to carry out a decision. Suppose a government puts a model supplied from abroad to work reviewing public contracts. Once staff have built the review around the system, losing access means finding another way to handle the contracts that are waiting, with the same records and checks. The people waiting for an answer will look to their own government, even if the company that can restore access has to answer to a different one. [13]
When I look back at printing, radio and the internet, I’m struck by how much changed once people had a new way to reach each other. Over time, it changed how they organized and what governments had to pay attention to. With AI, I wonder what happens when a handful of people can do work that used to take a much larger organization.
How much influence could a group that small have over what a government does, and what would it take to hold on to that influence? If this becomes part of how society works, what changes about the way we organize ourselves and the way governments respond to us? Would more people have a say in the decisions that affect them, or would a few people gain power the rest of us have little way to check? And how would we tell which was happening?
I have a lot more in my head on this one.
Harry Glorikian is the author of The Invisible Interface: How AI Turns Intentions Into Actions – And Who Wins (Ideapress Publishing / Simon & Schuster, June 2026). He is General Partner at Scientia Ventures, an Affiliate Researcher at the MIT Media Lab, and host of The Harry Glorikian Show.
References
[1] Team Mirai. Transcript of the party’s manifesto announcement. July 7, 2025.
[2] House of Councillors of Japan. Member profile of Takahiro Anno.
[3] House of Representatives of Japan. Membership by parliamentary group. September 28, 2026.
[4] Library of Congress. Thomas Paine’s Common Sense, 1776.
[6] New York State Office of the Attorney General. Fake Comments. May 2021.
[7] Hackenburg et al. AI can persuade people to take political actions. PNAS, October 2026, Study 2.
[8] Federal Communications Commission. Forfeiture Order FCC 24-104. September 30, 2024.
[9] Dragos. AI-Assisted Compromise of Mexican Water Utility with OT Implications. Updated May 2026.
[11] Anthropic. Statement on the directive to suspend Fable 5 access. June 12, 2026.
[12] Anthropic. Redeploying Claude Fable 5. June 30, 2026, updated July 1.
[13] Harry Glorikian. Who Owns the Operating Layer? May 21, 2026.

