This video of Republican Representative Thomas Massie having dinner with two Democratic colleagues was shared widely during Kentucky’s heated GOP primary.
Something doesn’t look quite right? That’s because it isn’t.
Massie’s primary opponent created the ad using generative AI. Many older constituents believed it. In fact, Massie blames the ad for losing the senior citizen vote in his district for the first time in his 14-year congressional career.
Generative AI is being used in political campaigns across the nation. It’s getting harder to discern what’s real. And that’s only part of the problem.
Welcome to the
‘You Can’t Trust Your Own Eyes’
Election
AI slop is taking over political communication. What could go wrong?
Artificial intelligence has become ubiquitous, inexpensive and convincing.
But when the public believes that vast amounts of political speech may be AI-generated, people start to distrust everything they see and voters could disengage from the democratic system. We need more safeguards.
There are three types of AI-generated political content: those that use satire to entertain or trigger you emotionally, those that are obviously fake but reinforce a negative narrative, and those that try to deceive.
There’s the satirical ad depicting James Talarico, the Texas Democratic candidate for Senate, singing a rewrite of “My Favorite Things,” with new lyrics targeting progressive and transgender policies. Or the one with former reality-TV star Spencer Pratt, which generated millions of views during his losing campaign against incumbent Los Angeles Mayor Karen Bass. Or the one fabricated by New York’s Republican candidate for governor, Bruce Blakeman, to target incumbent Kathy Hochul.
Each of these ads is entirely fake and was intended to be satirical, but together they have amassed more than 5 million views. All are still visible on X.
Some fakes are so sophisticated they’re almost impossible to detect. Consider the of former President Joe Biden urging people not to vote in the 2024 New Hampshire primary, created by a consultant working for Democrat Dean Phillips' campaign. When Bloomberg News ran it through a “speech classifier” tool developed by AI startup ElevenLabs, it was found to have only a 2% likelihood of being synthetic. Days later, a voice-fraud detection company determined that the technology used to make the deepfake audio was created by ElevenLabs.
If the AI companies can’t detect the fake content that their own technology creates, the public surely can’t. (ElevenLabs says they’ve since added new safeguards.)
All of this gives the bad actors a twofer — what scholars call the “liar’s dividend.” The more political communication becomes saturated with synthetic images and audio, the more likely the public will dismiss all political messaging as lies. Such distrust could lead to a cascade of harms, from increasing polarization to declining rates of participation in our democracy.
This shouldn’t be a partisan issue. Although media reports suggest that Republicans have so far leaned into AI-enhanced ads more than Democrats, members of both parties have deployed them. Democratic congressional candidate Blake Gendebien released a satirical attack ad against Republican incumbent Elise Stefanik. And AI ads have featured on both sides of this year’s Texas Senate primary. Democrat Jasmine Crockett was accused of using AI to enhance the size of the crowd in one of her ads, while the Republican candidates, Senator John Cornyn and Texas Attorney General Ken Paxton, traded satirical smears.
And fake attack ads aren’t the only reason to be concerned about AI’s influence on our election.
Oren Etzioni, a computer scientist who has been building AI systems since the late 1980s, founded TrueMedia.org to build tools for newsrooms and fact-checkers to detect deepfakes in the 2024 election. This year, his doomsday scenario is what he calls the “denial of democracy attack.”
Instead of a single, viral deepfake, thousands of AI-generated bots swarm hundreds of elections across the country — from Congress and state offices to school boards and mayoral races — with coordinated fake media at a scale that current detection systems can’t handle. Such AI swarms could exploit online echo chambers to manipulate public opinion, persuading communities of a threat or a consensus that is totally fictitious.
Some states are trying to prevent this and other doomsday predictions. Thirty-one states regulate political deepfakes. Of those, only three ban deepfakes in some capacity:
Regulation of Political Deepfakes Falls on States
Nineteen states do not have disclosure laws around AI use in political campaign ads
Source: NCSL
But we need federal standards. A bipartisan group of senators has advanced the NO FAKES Act, which would create new protections against AI-enabled replicas and deepfakes of someone’s name, image and likeness and allow people to sue over the unauthorized use of “computer-generated recreations.” Such a law might have blocked the AI-generated attack ad that was so effective against Massie.
As the US heads into the midterm elections, a perfect storm of AI-generated misinformation is gathering. Confronted by increasingly sophisticated technology, actual humans — regulators, journalists, candidates — are struggling to catch up.
All political communications — not just campaign ads — that are generated with AI should be required to be disclosed or be subject to serious fines. Penalties should be levied on organizations that benefit from AI-assisted impersonation of political candidates. Candidates should have to assume more responsibility when deception is used on their behalf. Tech companies should be compelled to offer more and better tools for candidates and elected officials, such as YouTube’s likeness-detection tool. Such a law would also apply to President Donald Trump’s fetish for posting AI content depicting his critics — and to California Governor Gavin Newsom’s attempt to troll the president using the same tactics.
In addition to more AI transparency and raising the stakes for bad actors, we also need to teach citizens to think critically. In this, Finland provides a model the US could emulate. Russia has swamped its Nordic neighbor with misinformation for a century, but when Moscow invaded Crimea in 2014, its tactics moved online. Finland made arming the public against false information a civic goal and integrated media literacy into its schools. Now, as deepfakes become widespread, Finland has added AI literacy to its school curriculum. Children as young as 3 are taught to “stop, think, check.” It’s no coincidence that Finns rank highly in both media literacy and trust in journalism.
Etzioni’s advice: “When you see something that galls you, which leaves you with an emotional reaction, that’s a great time to slow down before sharing it and ask yourself if this is real. What is the source? Is it corroborated?”
Another option is obvious but remarkably low on a campaign’s priorities list: Candidates should rely more on direct contact with voters and less on digital relationships.
That’s right — to prevent falsehoods from undermining our democracy in this age of technological abundance, we must do the hard work of engaging with people.
Added comment from ElevenLabs to the seventh paragraph of the article published Sept. 1.