AI Deepfakes Are Already Flooding the 2026 Elections
Subtitle: Fake political videos are running right now in real races. And most states still have no rules about it.
Key Takeaways
- AI-generated political ads are already running in 2026 Texas primary races, with candidates using deepfakes to attack opponents.
- A political consultant who sent AI-generated robocalls impersonating President Biden was hit with a $6 million fine and 26 criminal charges.
- Only 28 states have laws addressing political deepfakes, and most just require a disclosure label.
- California’s attempt to ban deceptive deepfakes in elections was struck down as unconstitutional.
- New technology called Content Credentials could help, but adoption is still early.
The 2026 midterm elections are already underway in several states. And something new is happening in this cycle that should worry every voter: AI-generated fake videos of real candidates are running as political ads, and in most of the country, there’s nothing illegal about it.
This isn’t hypothetical. It’s already happening in Texas.
What’s Happening in Texas Right Now
Texas held its primary elections in March 2026, and the campaign trail was loaded with AI-generated content.
Senator John Cornyn posted an AI-generated video depicting U.S. Rep. Wesley Hunt as a “show dog.” The video was sophisticated enough to fool casual viewers, and it carried no disclosure that it was AI-generated.
Attorney General Ken Paxton shared an AI video showing his Republican primary opponent Senator Cornyn dancing with Democratic Rep. Jasmine Crockett. Paxton used the fake video to claim Cornyn was “dancing late into the night with liberal lunatics,” attacking him for bipartisan work on the opioid crisis.
Rep. Jasmine Crockett posted an AI video featuring herself as a baby accusing “Baby Trump” of rigging Texas elections. It came from a YouTube channel of political parodies called Diaper Diplomacy.
And in a down-ballot race, GOP candidate Kat Wall published an AI-generated YouTube video against her opponent Rep. Angelia Orr, featuring deepfakes of Vladimir Putin and Xi Jinping with synthetic voice clones to mock Orr’s voting record and frame her as an ally to foreign adversaries.
None of these ads were required to disclose that they used AI. A Texas bill that would have mandated disclosure passed the state House but stalled in the Senate. So AI in political advertising remains a free-for-all.
The Precedent That Should Have Changed Everything
You’d think the political world would have learned from what happened in New Hampshire.
In January 2024, thousands of New Hampshire voters received robocalls from what sounded exactly like President Biden telling them not to vote in the upcoming primary. “What a bunch of malarkey,” the fake Biden said, using his signature phrase. The message falsely claimed that voting in the primary would prevent them from casting ballots in November.
The call was traced back to political consultant Steven Kramer, who admitted to orchestrating the scheme. He used commercially available AI voice-cloning tools to generate the fake Biden audio, then hired a telecom company called Lingo Telecom to blast it to voters.
The consequences were severe. The FCC hit Kramer with a $6 million fine, its first penalty involving generative AI. Lingo Telecom faced a separate $2 million fine. Kramer was charged with 13 felony counts of voter suppression and 13 misdemeanor counts of impersonation of a candidate across four New Hampshire counties.
The FCC then banned AI-generated voices in robocalls entirely.
But here’s what matters: the New Hampshire case was a robocall. What about video ads posted on social media? What about AI-generated images shared on campaign websites? What about deepfake clips that go viral on platforms where no one checks whether they’re real?
That’s where the law falls apart.
The Legal Patchwork
As of early 2026, 28 states have enacted laws that specifically address deepfakes in political communications. But most of them only require disclosure, a label saying something like “This ad was generated or substantially altered using artificial intelligence.”
That’s it. Not a ban on deceptive content. Just a label.
And even that modest approach is hitting legal walls. California passed AB 2839 in September 2024, which tried to prohibit deceptive deepfakes in elections. A federal judge struck down key portions in August 2025, ruling that the restrictions conflicted with the Communications Decency Act and were likely unconstitutional under the First Amendment.
The ruling exposed a fundamental problem: political speech is the most protected category under the First Amendment. Courts have historically been extremely reluctant to let the government decide what’s “deceptive” in a political context. That makes regulating AI-generated political content far harder than regulating it in other domains like pornography or commercial fraud.
At the federal level, the picture is only slightly better. The TAKE IT DOWN Act, signed in May 2025, provides the first nationwide framework for intimate deepfakes but doesn’t cover political ones. The FCC’s robocall ban is specific to voice calls. There’s no federal law that specifically addresses AI-generated video or images in political advertising.
Why This Is Different From Normal Political Lying
Politicians have always stretched the truth. Opposition research has always been aggressive. Attack ads have always been misleading. So why is AI different?
Three reasons.
Speed and scale. Creating a convincing fake video of a political opponent used to require Hollywood-level skills. Now it takes a laptop and a few hours. The tools are widely available, and the quality is getting better every month. What was obviously fake two years ago might fool most people today.
Plausible deniability. When a candidate posts an obviously manipulated image, everyone understands it’s satire or exaggeration. But AI-generated content sits in an uncanny valley where it’s realistic enough to confuse people but can always be defended as “just a joke” or “clearly AI.” Paxton’s dancing video of Cornyn is a perfect example: fake enough to be defensible, real enough to mislead.
The asymmetry of correction. A fake video can reach millions of people in hours. The correction, if it comes at all, reaches a fraction of that audience. Research consistently shows that people who see misinformation retain the false impression even after seeing a correction. In the context of an election, where every vote matters and timing is everything, that asymmetry is devastating.
Can Technology Fix What Technology Broke?
The tech industry is working on solutions. The most promising is called Content Credentials, developed by the Coalition for Content Provenance and Authenticity (C2PA). Major companies including Adobe, Microsoft, Google, and OpenAI are backing it.
Here’s how it works: when a photo, video, or audio file is created, the tool embeds a cryptographic signature, essentially a digital birth certificate. This signature records who created the content, when it was created, and what tools were used. If someone edits the content, the edit history is preserved. If the content is AI-generated, that fact is recorded.
The system also uses invisible watermarks and content fingerprints that survive even if someone strips the visible metadata. So even if someone downloads a video, removes the credits, and re-uploads it, the system can often still trace it back to its origin.
But Content Credentials have a big limitation: they only work if the creator opts in. A politician who intentionally creates a deepfake to mislead voters isn’t going to voluntarily tag it with a “this is AI-generated” label. The technology is better suited for proving that legitimate content is real than for catching fakes.
California’s SB 942 tries to bridge this gap by requiring large AI platforms to provide free detection tools and include watermarks in their output. But it doesn’t take effect until August 2026, months after several primary elections will have already taken place.
What Voters Can Actually Do
The regulatory framework isn’t going to save you in 2026. The laws are too slow, too fragmented, and too easily challenged. So what can you do?
Be skeptical of viral political content. If a video of a politician saying something outrageous is circulating on social media without a source from a major news outlet, treat it with suspicion. Check whether real journalists have verified it before sharing.
Look for Content Credentials. Some platforms are starting to display provenance information on images and videos. If you see a “Content Credentials” or “CR” icon, click it. It will show you the history of that piece of content.
Check the context. AI deepfakes often circulate without context, as standalone clips without the surrounding speech, event, or press conference. If you can’t find the full event that a clip supposedly came from, that’s a red flag.
Report suspected deepfakes. Most social media platforms have reporting mechanisms for manipulated media. Using them won’t instantly remove content, but it feeds into moderation systems that can flag the content for other users.
Follow the money. Real political ads have to disclose who paid for them. If a political video has no “paid for by” disclaimer, it’s either not an official ad (which means less accountability) or it’s breaking existing campaign finance rules.
Where This Is Heading
The 2026 midterms are going to be the first major U.S. election cycle where AI-generated political content is widely used. Not as a novelty. Not as an experiment. As a standard campaign tactic.
The technology will get cheaper and more convincing. The legal framework will continue to lag behind. And voters will be left to navigate an information environment where seeing is no longer believing.
The question isn’t whether AI will be used to mislead voters this year. It already is. The question is whether we’ll build the tools, laws, and habits to deal with it before the damage is done.
Based on what’s happening in Texas right now, we’re not there yet.
This article is part of Winging AI’s ongoing coverage of how artificial intelligence affects everyday life. We research and write about AI so you don’t have to decode the hype yourself.