How AI-Forged Disinformation Is Hijacking Japanese Elections

The votes are counted in Okinawa. But the real battle happened somewhere else entirely.

The Okinawa gubernatorial election, decided on September 13, produced a classic face-off between Governor玉城デニー, the incumbent, and former Naha deputy mayor古謝玄太 as the sole serious challenger. Four other newcomers shared the ballot. What made this contest notable — and potentially dangerous for Japanese democracy — was not the policy debate. It was what happened on social media.

Local newspaper Ryukyu Shimpo analyzed 2,871 YouTube videos uploaded in the week after the official campaign announcement. Of the top 30 most-viewed videos, 18 had verifiable origins outside Okinawa Prefecture. The remaining 12 were likely external as well. Zero could be confirmed as local. These top 30 videos accounted for roughly 35 percent of the roughly 50.2 million total views on Okinawa-related content. And 20 of those 30 carried negative messages about the incumbent.

That is the single most striking finding: when it comes to Okinawa’s election, the loudest voices on the internet are almost certainly not from Okinawa.

The accounts are familiar

The pattern from this race is not new. It mirrors what happened in Hyogo Prefecture’s gubernatorial election in 2024 and Miyagi Prefecture’s in 2025, both flagged by Kobe Shimbun data. In Hyogo, total video views exceeded 180 million across 18 days — inflated in part by out-of-prefecture influencers riding the coattails of a high-profile race.

A post-election survey of 1,991 Hyogo residents found that 21 percent said social media and 19 percent said YouTube were the sources they saw most often, compared with 32 percent who cited newspapers and newspaper sites. A significant minority — roughly 40 percent — had their political information diet shaped primarily by platforms whose algorithms reward engagement over accuracy.

Several accounts that topped the Okinawa charts also posted heavily during Hyogo’s campaign. The number one video in Okinawa — posted by the channel “おみそちゃんねる【世界どん深闇ニュース】” — accumulated roughly 4.9 million views in its first week. It criticized Okinawa’s base opposition movement. Before Hyogo’s election, the same account ran more than 30 videos about the region, each exceeding 100,000 views, including one claiming that a governor who tried to reform Kobe Port’s interests provoked violent backlash from vested groups and media crusades. That claim was inaccurate.

A second top video — “真相深入り！虎ノ門ニュース” featuring journalist Shinya Suda — also pushed substantial Okinawa content, including an interview with candidate Takashi Tachibana, currently on trial, who had campaigned under the “二馬力選挙” banner.

The same accounts, the same format, the same economic incentive. Outside influencers treat Japanese prefectural elections as scalable content farms.

X tells a sharper story

YouTube data is useful, but X — formerly Twitter — reveals the mechanism more clearly. Using Users Local’s SNS analysis tool “Social Insight,” Kobe Shimbun found that as of September 10, roughly 1.507 million posts referenced the Okinawa gubernatorial election. About 90 percent originated outside the prefecture.

Among those posts were images styled like campaign posters for a new candidate — forged. There were also videos defaming the incumbent, several of which appear to have been AI-generated or AI-enhanced. Similar patterns appeared in Hyogo (about 80 percent out-of-prefecture across roughly 741,000 posts) and Miyagi (roughly 80 percent out-of-prefecture across nearly 200,000 posts).

This is no longer a theoretical problem. AI can fabricate convincing campaign posters, synthetic videos, and deepfake audio at near-zero marginal cost. An operator in Osaka can target Okinawa voters without ever visiting the island. The geometry of the threat is simple: anyone with an account and a prompt can participate.

Who wins, who loses

The immediate beneficiary of this ecosystem is the content creator. YouTube monetizes views. X monetizes attention. An influencer who cross-posts fabricated or exaggerated election content across multiple prefectures maximizes returns while minimizing risk — because the content rarely ties to any single local community that could hold them accountable.

The loser is the local voter. The Kobe survey showed that nearly half of respondents in Hyogo encountered more election information via social platforms than via traditional press. If the majority of that information is curated by outsiders with no skin in the game, the electorate’s understanding of the issues distorts. Base policy, fiscal management, and candidate backgrounds get displaced by drama, controversy, and manufactured outrage.

The incumbent in any given prefecture also bears disproportionate risk. Anti-incumbent videos dominated the Okinawa top 30. Out-of-prefecture actors have less incentive to protect the status quo and more incentive to amplify friction. A governor known for taking hard lines on U.S. base issues — like玉城デニー — is a natural magnet for sensational content. That content does not need to be true. It needs only to be shareable.

What happens next

Okinawa’s two major newspapers, Ryukyu Shimpo and Okinawa Times, began fact-checking campaigns after the September 27 announcement. That is a step in the right direction, but it is reactive. Fact-checking arrives after a video has already accumulated millions of views. By the time the correction circulates, the original distortion has set.

Platforms themselves have shown little appetite for proactive intervention in Japanese local elections. X does not label or suppress synthetic media at scale. YouTube’s partner policies target fraud and spam but do not address the core incentive structure: viral election content earns revenue regardless of origin or accuracy.

The Hyogo survey should be a warning. If 40 percent of voters there were getting their primary election information from social platforms, and 80 percent of those posts were external, then Japanese prefectural elections are increasingly contests fought on terrain controlled by non-local actors. The trend is accelerating as generative AI tools become cheaper and more accessible.

Three questions deserve urgent answers. First, can platform operators be compelled to label or demote content that appears AI-generated in the context of active elections? Second, should election commissions track and publish the geographic origin of high-reach political content? Third, and most difficult, how do you regulate an actor in Kobe posting about Okinawa without opening the door to broader speech restrictions?

There is no clean answer. But the Okinawa election has already demonstrated the blueprint: out-of-prefecture influencers, amplified by AI tools, flooding local races with synthetic content designed for maximum reach rather than maximum truth. The next race will not look different. It will look louder.