AI Deepfakes Are Flooding the 2026 Midterms. 26 States Scramble

TL;DR: AI-generated deepfakes showing candidates saying things they never said are already circulating in 2026 campaigns. A deepfake of Jon Ossoff ran in Georgia's Senate race. Andrew Cuomo's campaign posted (then deleted) a racist AI attack ad. 26 states now have deepfake election laws, up from 5 in 2023, but the FEC has done nothing. Most state laws only require disclosure labels, not bans. You can't trust your eyes anymore. Here's what voters need to know.

It's Already Happening

Forget "could happen" or "might happen." Deepfakes are live in 2026 campaigns right now.

In November 2025, Rep. Mike Collins' campaign released an AI-generated video of Senator Jon Ossoff in Georgia's Senate race. The deepfake shows a computer-generated Ossoff mocking farmers, saying he'd "only seen a farm on Instagram" and defending a government shutdown. None of it real.^[1]

Collins' campaign slapped a tiny disclaimer on the screen saying "this video is AI-generated" (just enough to technically comply with disclosure laws) and kept running it. They say they'll keep using AI tools for voter outreach.

Then there's Andrew Cuomo. In late October, his campaign posted an AI-generated attack ad titled "Criminals for Zohran Mamdani" during the NYC mayoral race. The video showed AI-generated "criminals" including a Black man in a keffiyeh shoplifting and a synthetic pimp announcing support for his opponent. The campaign deleted it within minutes and blamed a "junior staffer."^[2]

This isn't theoretical. It's the 2026 election cycle.

26 States Have Laws. Most Won't Stop Much.

State legislatures have moved faster than Congress. From 5 states with deepfake election laws in 2023, we're now at 26:^[3]

Alabama, Arizona, California, Colorado, Delaware, Florida
Hawaii, Idaho, Indiana, Kentucky, Massachusetts, Michigan
Minnesota, Mississippi, Montana, Nevada, New Hampshire, New Mexico
New York, North Dakota, Oregon, Pennsylvania, Rhode Island, South Dakota, Texas, Utah, Washington

Five more states (New Jersey, Virginia, Maryland, Tennessee, and Vermont) are considering bills right now.

But here's the catch: most laws only require disclosure, not bans. Stick a label on your deepfake, even a barely visible one, and you're technically legal. Texas goes further, prohibiting deepfake videos intended to influence voters within 30 days of elections.^[4] Maryland criminalizes AI deepfakes for election misinformation under SB0141.

California tried to ban deceptive deepfakes outright. A federal court struck it down as a First Amendment violation.^[5] That's the constitutional wall other states keep hitting.

The FEC Has Done Nothing

While states scramble, the Federal Election Commission has been deadlocked.

Back in August 2023, the FEC asked for public comments on AI in campaign ads. Since then? Interpretive guidance saying existing fraud laws apply "regardless of technology." That's it. No new rules. No specific requirements.^[6]

The FEC, working with DOJ and FTC, is supposedly drafting rules for AI content ethics and synthetic media, expected "by mid-2026." So after the primaries are underway. After deepfakes have already circulated.

The FCC at least did something. In February 2024, they ruled that AI-generated voice clones count as "artificial or prerecorded voices" under the Telephone Consumer Protection Act, making deepfake robocalls illegal without prior consent.^[7] That came after voters in New Hampshire received deepfake calls impersonating President Biden designed to suppress turnout.

Detection Doesn't Work Like You Think

Bad news: you probably can't spot a good deepfake.

The telltale signs everyone learned about (weird hands, uncanny eye movements, mismatched shadows) are increasingly outdated. Modern AI generators fix those artifacts. Looking for "tells" is like checking for fakery by looking for obvious Photoshop edges. The technology moved on.^[8]

Automated detection tools aren't reliable either. They're trained on yesterday's deepfakes. Today's generation already evades them. The Reuters Institute found that AI detection tools have limited accuracy and "wax and wane as AI-generation tools become more sophisticated."^[9]

What actually works:

Check the Source

Did the candidate's official campaign post this? Is it on a verified account? If you found it reshared on social media without clear origin, be suspicious.

Listen for Out-of-Character Claims

If a candidate suddenly says something wildly inconsistent with their established positions, especially something explosive, verify with official campaign channels or journalists covering the race.

Slow Down

The purpose of election deepfakes is to go viral before verification. Taking even 30 seconds to check before sharing denies the creators what they want.

Use Multiple Media

Studies show people detect deepfakes better when audio and subtitles are present: more channels means more opportunities for inconsistencies to surface.

Most Campaigns Have No Plan

Here's something that should worry you: when NOTUS asked lawmakers about emergency protocols for deepfake attacks, the answers were grim.^[10]

Rep. Obernolte: Not aware of any campaign protocols. Rep. McBride: Same. Rep. Burchett, who has already had deepfakes of himself circulating online, was blunt: "None whatsoever."

Campaigns are flying blind. Political strategists recommend maintaining consistent authentic social media presence, establishing rapid-response channels, and building relationships with platforms before crises hit. Most campaigns haven't done this work.

The asymmetry is brutal: creating a deepfake takes hours. Debunking it and repairing the damage takes days or weeks. And in a close race, that window matters.

What's Coming

The 2026 midterms feature high-stakes gubernatorial races and a competitive Senate map. Experts at the R Street Institute warn that while "meaningful impact on actual election results is unlikely," deepfakes will "poison the information environment."^[5]

International precedent is concerning. In Argentina's May 2025 elections, deepfakes falsely claiming a candidate withdrew appeared hours before polls opened. In Ireland, a similar deepfake dropped four days before their presidential election. In Romania, deepfake videos promoted fake government investment schemes using presidential candidates' faces.^[11]

The playbook is clear: drop synthetic content too late for effective debunking, target close races, create chaos.

What You Can Do

Verify Before Sharing

That viral clip might be synthetic. Check if major news outlets are covering the same claims. No coverage? That's a red flag.

Follow Official Channels

Candidates' verified accounts, campaign websites, and reputable journalists covering the race. If something explosive isn't on those channels, question it.

Report Suspicious Content

Platforms do take down deepfakes, especially when reported quickly. Report to the platform, and if it's about a federal election, to the FBI's election crimes coordinator.

Know Your State's Law

In Texas, unlabeled deepfakes within 30 days of an election are illegal. In your state? Check Public Citizen's tracker to know what protections exist, and what gaps remain.

The Bottom Line

The 2024 election dodged the worst predictions about AI-generated chaos. The 2026 midterms won't be so lucky.

The technology is better. The tactics are proven. And the regulatory response (26 state disclosure laws and a silent FEC) isn't close to matching the threat.

We're running the 2026 elections on an honor system where bad actors can manufacture reality and face minimal consequences. The deepfakes are already here. The primaries are months away. And every voter is now responsible for something that used to be obvious: figuring out what's real.