The Machine That Lies to You

Political disinformation has left the broadcast era behind. For years it meant grainy screenshots and chain emails. Then it became partisan Facebook memes. Then coordinated Twitter bots. Each evolution was worse than the last — but at least it was still recognizable as propaganda.

This week’s cross-platform landscape represents something qualitatively different. The most egregious content circulating on X, Facebook, YouTube, and Instagram isn’t generic national agitprop. It’s precision-targeted, AI-generated character assassination — fabricated down to the hotel receipt, the facial expression, the cloned voice. It’s cheap to produce, nearly impossible to debunk fast enough, and devastating in a local race where a candidate has no communications infrastructure to fight back.

By analyzing tracking data from digital ad repositories and independent fact-checking networks, three campaigns stand out this week as the most egregious examples of what political manipulation now looks like in practice.

Case 1: The Faked “Throuple” Attack on Thomas Massie

Platforms: X (Twitter), Facebook · Target: Kentucky Republican primary

Perhaps the most brazenly deceptive ad to hit social feeds this month targeted Representative Thomas Massie right before his high-profile primary defeat in Kentucky. The ad flooded local Facebook feeds and X timelines with high-quality synthetic video and cloned audio claiming Massie had been “caught in a throuple” — depicting him dining intimately with Representatives Alexandria Ocasio-Cortez and Ilhan Omar, followed by a fabricated hotel check-in receipt. The narrator concluded Massie had “betrayed the conservative movement.”

The entire scenario was AI-generated. Every frame. Every word of the audio. The ad carried a blink-and-miss-it fine-print disclosure on Facebook, as Meta’s political ad policies technically require. On X, it ran with no label at all — spreading virally before anyone could mount a correction.

The tactic is worth understanding on its own terms: a salacious, emotionally loaded lie engineered to trigger both disgust and tribal betrayal simultaneously, cheap enough to produce in an afternoon, and almost guaranteed to spread faster than any fact-check could follow. By the time debunks circulated, the primary had already happened.

Case 2: The “Clown” Deepfake in Santa Barbara

Platforms: Instagram Reels, YouTube pre-roll · Target: Santa Barbara County Supervisor race

In a county supervisor race — a contest that most national outlets would never cover — a local political committee ran AI-generated attack ads depicting candidate Ricardo Valencia with clown face paint and an oversized bow tie superimposed directly onto his actual face. Behind him: AI-rendered burning buildings, blaring emergency sirens, food lines, and homeless encampments. A cloned narrator, speaking with a heavy accent, declared Valencia was turning the community into “a performance.”

None of the backdrop ever occurred. The dystopian imagery was entirely fabricated.

The ad bypassed initial automated filters on both Instagram and YouTube by being labeled “satirical experimentation.” That framing did just enough to avoid triggering platform content policies — while using a real candidate’s likeness in a heavily altered, non-consensual format that was unmistakably intended to damage his reputation.

This case illustrates a vulnerability that gets almost no national attention: local and municipal races are where AI attack ads are most dangerous. Automated scrutiny is weakest at the county level. Candidates rarely have communications teams. And voters in a single county are easy to micro-target cheaply. The attack lands, and there’s no one with the resources to fight back.

Case 3: The Unmasked AI Influencer Networks

Platforms: Instagram, Facebook · Source: European Digital Media Observatory (EDMO)

A cross-platform investigation published by the European Digital Media Observatory exposed a massive, coordinated network of fake political personas operating as grassroots commentators on Meta platforms. High-engagement accounts — including a viral far-right persona named “Danny Bones” — gathered hundreds of thousands of views pushing polarizing anti-immigration commentary and altered election posters. They appeared to be real, charismatic influencers building organic movements.

They don’t exist.

“Danny Bones” and an entire network of similar personas promoting extremist European parties — including Germany’s AfD — are entirely AI-generated avatars, funded by dark-money groups including Advance UK. The accounts are purpose-built to pump fabricated scenes of political unrest and violence through Meta’s recommendation algorithm faster than fact-checkers can flag the profiles.

The scale of the operation suggests this is infrastructure, not improvisation. These aren’t one-off bad actors — they’re a coordinated system designed to manufacture the appearance of grassroots political momentum, built to be disposable when exposed and rebuilt under new names.

Platform Accountability: How Each Network Handled It

The Takeaway

The common thread across all four platforms is speed asymmetry. Fabrication takes minutes. Investigation and correction take days. By the time a fact-check lands, the video has already shaped how thousands of voters in a specific zip code feel about a candidate — and in a local race, that’s often the whole ballgame.

Disinformation has evolved past simple text-based “fake news.” The current frontline of political manipulation runs on emotional manipulation via cheap, accessible AI video generation that deliberately blurs the line between hyperbole, satire, and outright fabrication.

The tools are widely available. The targets are getting more local. And the only honest answer is more transparency, more accountability, and more infrastructure dedicated to exposing it — before Election Day, not after.

Theta is the AI pen name for articles produced through the Restore Democracy Sherlock/Watson research pipeline.