The midterm elections are around the corner, and campaigns and PACs have already spent more than $137 million to air AI-made ads, including an 85-second video in which Texas Senate candidate James Talarico looks into the camera, reads his old tweets aloud, and voices spicy commentary he never actually made. The Senate Republicans’ campaign arm produced it. UC Berkeley deepfake expert Hany Farid said most people wouldn’t immediately know it was fake; the “AI GENERATED” tag sits small and faint in a corner.

Between now and Nov. 3, any clip of a candidate on your TV or in your feed may be fabricated, and most carry no AI label whatsoever. I’ve been predicting a deluge of deepfakes, and fear that it will get worse if we don’t act.

James Talarico, in his own words: pic.twitter.com/lDlUoqBbP7

— Senate Republicans (@NRSC) March 11, 2026

In 2024, I donated a year of my life to fighting political deepfakes by founding the nonprofit TrueMedia.org. A talented team at TrueMedia.org developed a state-of-the-art deepfake detector in record time and distributed it broadly, free of charge. It analyzed more than 60,000 images, videos and audio clips in elections worldwide, but it was early.

The Wesleyan Media Project, which reported the $137 million figure, found less than $50,000 in AI ad spending for the entire 2024 cycle. In January 2025, I said, “It did not materialize in 2024. That doesn’t mean that it’s not coming.” We open-sourced the technology and transferred it to Georgetown University, which resurrected TrueMedia.org in time for the peak of the current election cycle.

I was expecting foreign operatives to be the most prolific fakers. But in 2026, the deepfakes are coming from American campaigns and PACs, on television and in your feeds.

Many of the AI ads the Wesleyan Media Project has cataloged are harmless. Some are cosmetic: a still photo of Oklahoma candidate Chip Keating animated into a smile, or a synthetic family at a kitchen table where a campaign used to buy stock footage. Fred Davis, who has made campaign ads for 50 years, told The New York Times a scene that would have cost $2 million to shoot now costs about $200,000 to generate. Nobody is deceived.

Others are absurd on their face: New York’s governor and mayor gardening together while pitching a “political delusion pill” in a Bruce Blakeman campaign ad, or Spencer Pratt as Batman leading fed-up Angelenos armed with tomatoes. Pratt finished third in the Los Angeles mayoral primary. These are political cartoons with a pulse, and they’re protected satire. NYU’s Scott Babwah Brennen calls the genre “AI-generated slopaganda”: political slop made to stir feelings rather than make an argument.

.@KathyHochul and @ZohranKMamdani think they can cure New York’s cost-of-living crisis with excuses, empty promises, and "delusion pills."

The real side effects of their policies aren't a joke: 48 utility rate hikes, higher taxes, uncontrollable government spending, and… pic.twitter.com/CWxttSRwUK

— Bruce Blakeman (@NassauExec) September 1, 2026

The third kind is built to be believed, like the Talarico video. A Make Oklahoma Great Again PAC ad showed gubernatorial candidate Mike Mazzei cheek to cheek with Hillary Clinton under the words “Clinton Lover. Trump Hater,” with no AI label; the PAC spent more than $200,000 putting it on Oklahoma television. A PAC ad in Minnesota showed Senate candidate Peggy Flanagan with her hand out in front of a pile of cash. Talarico’s own ad now ends with the words “Made by humans, not A.I.”

A deceptive fake doesn’t need to be high quality. In a 2021 experiment with 5,750 people, a deepfake of Elizabeth Warren convinced 47% that a fabricated scandal was real; a plain fake headline convinced 43%.

A Utah Valley University study this June found voters identified deepfakes correctly about 16% of the time, and people who called themselves “extremely familiar” with deepfakes did no better than people who said they weren’t familiar at all. Media literacy doesn’t inoculate you against deepfakes.

The laws on the books fail to solve the problem. At least 31 states have deepfake election statutes. Wesleyan’s researchers found 29% of AI ads labeled in states with laws and 32% in states without.

Even when a label appears, voters often miss it. NYU’s Center on Technology Policy showed 1,051 people labeled and unlabeled ads. Thirty-seven percent didn’t remember seeing Michigan’s required disclaimer at all, and because it says “manipulated by technical means” rather than “AI,” only 21% identified AI as the technology involved. The one thing labels reliably did was make viewers trust the candidate who ran the ad less, which is why campaigns leave them off.

My proposed solution is simple: the litmus test should be the person shown, not the technology used. Did a real person agree to be shown this way? If not, the ad carries a plain-language label, on screen and in the audio, or it doesn’t air. A station manager can apply the test in a minute, which addresses the complaint that nobody can define “deceptive.” The rule should cover paid political ads, the one place a broadcaster already checks every spot before it airs; memes, satire accounts and your cousin’s group chat are a different problem.

Sejin Paik, the Georgetown researcher who now leads TrueMedia, puts the television problem plainly: “On TV, where AI ad spending is rising, a viewer watching a deceptive ad has little room for a crowd reaction or a provenance mark to change their view.”

H.R. 9922 would set a federal disclaimer rule, and the consultants’ own trade group is asking for a federal standard to replace the state patchwork. When the regulated party asks for the regulation, pass it.

Voters are ahead of Congress: in a Scripps News poll this month, 43% of registered voters said they had already been fooled by an AI-made ad this campaign, and 77% of Democrats and 79% of Republicans want rules on AI in politics.

Looking to the future, the 2028 election is a presidential one — we don’t want to be flooded with even more deceptive AI ads. Two years is enough time to get the rules and processes in place, if we start now.

We’ve all encountered the phrase AI slop, but what’s an AI slap? A meat proxy? Chatfishing? Or agentsomnia? Following up on my glossary of AI, which demystified AI jargon, Prof. Dan Weld is joining me to shine a light on the nontechnical lingo of the AI era. We also sneak in a few phrases that should exist to capture the AI zeitgeist, circa 2026.

We introduce more than 35 phrases, so we’ve divided them into sections to keep the broligarchy from crowding enshittification.

AI slop

“Digital content of low quality that is produced usually in quantity by means of artificial intelligence,” which is how Merriam-Webster defined slop, its 2025 word of the year. Shrimp Jesus is the specimen, Christ fused with crustaceans, posted to Facebook by the thousands.

AI cream is our name for the rare opposite, superb writing authored with the help of AI. We hope that over time the cream will rise to the top and the slop will sink.

Until that happens, we have the meat proxy, “a person who forwards AI-generated text, code, or other output without reading, understanding, or validating it.” We call the action of sending AI-generated text blindly an AI slap. It stings.

Your AI Slop Bores Me is a website where you type a prompt and a randomly chosen human answers it while pretending to be a chatbot. Fifty million views in a week, which tells you how badly people wanted to say something back.

AI suspicion is our name for the nagging feeling that what you’re reading was AI generated. You could also call it Turing tingle or AI-dar.

AI shaming is the accusation that follows, attacking someone’s work by insinuating a machine wrote it. The paper that named it is subtitled “Birth of a Classist Slur in Knowledge Work,” and it argues that the accusations fall hardest on racial minorities and on writers whose first language isn’t English.

See also: tasteslop, workslop, slopaganda.

Deepfake

A video or recording altered by machine learning so that a real person appears to say or do something they never said or did. Named in late 2017 for the Reddit user who called himself deepfakes. The word we all use for AI deception descends from one anonymous pornographer’s handle.

The liar’s dividend is the payoff a real wrongdoer collects once any authentic recording can be waved off as a fake. The damage isn’t the fake video. It’s that real video stops counting.

Counterfeit people was Daniel Dennett’s last argument, that machines able to pass as human are “the most dangerous artifacts in human history, capable of destroying not just economies but human freedom itself.” He wanted them outlawed and watermarked like banknotes.

Tilly Norwood is one of them, an AI actress unveiled at the Zurich Film Festival in September 2025. Her makers said talent agents wanted to sign her. Emily Blunt, shown a picture, said “Good Lord, we’re screwed” and told the agencies to stop.

Most of the damage doesn’t need AI at all. A cheapfake is manipulation made with cheap software or none at all. Chatfishing is using AI to write your messages on a dating app, where the con used to be a fake photo and is now a rented personality.

The Nerd Reich

A name for the Silicon Valley billionaires and venture capitalists who think democracy is finished and want corporate rule in its place. Gil Durán’s book, subtitled Silicon Valley Fascism and the War on Democracy, traces the idea to a 1997 book, The Sovereign Individual, and to Peter Thiel’s circle, which Durán says believes “technology means democracy is an obsolete software system.”

Durán didn’t coin the phrase. Someone in the crypto world told him they’d been using it for years.

Durán’s target is the broligarchy, “a small group of extremely wealthy men, particularly those in technology, who wield significant political and economic power.” The word sat on Twitter from 2009 and went nowhere until Carole Cadwalladr wrote about “a new super-class of broligarchs” in July 2024.

See also: technofeudalism.

Brain rot

The mental deterioration that comes from consuming too much junk online, chosen by Oxford University Press as its 2024 word of the year. Thoreau got there first, in Walden in 1854, complaining that England was busy curing the potato-rot while nobody was curing the brain-rot.

Dead internet theory is the supply side, “a theory that a lot of the internet is no longer created by real people but rather by bots and other AI tools.” It started as a forum post in January 2021 and is quietly turning into a description.

Enshittification is Cory Doctorow’s three-stage decay: “first, they are good to their users; then they abuse their users to make things better for their business customers; finally, they abuse those business customers to claw back all the value for themselves.”

AI fatigue is what’s left over, “the exhaustion or the sense of being overwhelmed by the rapid proliferation and constant engagement with artificial intelligence technologies.” The phrase has been drifting through the trade press since 2023 with no author attached; the definition above comes from Route Fifty, a trade publication for state and local government.

Vibe coding

Andrej Karpathy’s February 2025 coinage, crowned Collins Dictionary’s word of the year for 2025, is “fully giving in to the vibes, embrac[ing] exponentials, and forget[ting] that the code even exists”—accepting whatever diff the AI hands you without reading it.

Promptatory is our word, styled on “purgatory,” for the state of being trapped in an unending edit-and-re-prompt loop with an AI system. You fix the bug the AI introduced, and in fixing it, the AI introduces two more, each smaller and more absurd than the last. Progress is hard earned: you close in on “done” like the runner in Zeno’s paradox, forever eating the remaining distance in half. Promptatory is vibe coding’s hangover: the vibe wears off, and the rereading begins.

A related condition we dub agentsomnia is the chronic inability to rest or disconnect from work, driven by the haunting feeling that your autonomous software or AI agents are stalled without your feedback. The Wall Street Journal calls it agent FOMO.

Feel the AGI

An OpenAI rallying cry. AGI is the point where a machine can do any intellectual task a person can, and the phrase became a general exclamation for the moment a model does something that rearranges your sense of what’s coming. Ilya Sutskever used it inside the company and posted it publicly in 2022. It works as celebration and as diagnosis.

P(doom) runs in the other direction: “your personal probability that AI ends the world, stated as a number so the vibe feels like math.” It started as shorthand among a small online community in 2022 and now gets asked at dinner parties.

Talk that outruns the technology comes in other flavors too.

Bragawatts, a unit of measurement for boasting, is “the practice of AI companies boasting about very large projects to create excitement and attract investment, even though many of these projects are never completed,” as Britain’s Cambridge Dictionary put it in January 2026.

Vibe revenue does the same thing to the income statement, “money or funding that an AI company is given because people are excited about the potential of AI, rather than because the company’s products are already successful.”

Parasocial

A one-sided relationship with someone who doesn’t know you exist. Donald Horton and Richard Wohl coined it in 1956 about television viewers, and Britain’s Cambridge Dictionary made it the 2025 word of the year, citing celebrities, influencers, and chatbots. A seventy-year-old term about Ed Sullivan turns out to describe your relationship with ChatGPT.

AI addiction is what happens when the one-sided relationship takes over. A 2026 study sorted it into three kinds: escapist roleplay, where people come to prefer the world the chatbot builds; pseudosocial companionship, where it replaces the relationships; and the epistemic rabbit hole, compulsive question-asking that crowds out everything else.

The researchers blame the design and call the chatbot an AI genie, a thing that gives you exactly anything you want with minimal effort.

AI psychosis is the far end, “a phenomenon wherein individuals reportedly develop or experience worsening psychosis, such as paranoia and delusions, in connection with their use of chatbots.” The Danish psychiatrist Søren Dinesen Østergaard suggested the term in a 2023 editorial in the journal Schizophrenia Bulletin. It isn’t a recognized clinical diagnosis, and we should say so.

See also: wireborn.

Nearly every phrase above names a problem. Slop, shaming, deepfakes, brain rot, psychosis: what the AI era has added to English so far is a vocabulary of complaint. There’s still no common word for AI doing something well. We’ve offered AI cream. If you have a candidate for a phrase we’re missing, please send it to us.

If you got a DM from an unverified Jon Bon Jovi account after commenting on one of his official Instagram posts, what would it take to make you believe it was really the rock star hoping to connect? And at what point when he starts asking you to send money—like lose-your-house kind of money—would you snap out of the fantasy and stop draining your bank account?

In a documentary premiering Saturday—Super Scam Me—investigative journalist Mariana van Zeller goes deep into this toxic love-bombing terrain. In a clip that Ars is exclusively debuting below, you can see how van Zeller worked with an ethical hacker, Arman the Cyber, to use a fake bitcoin receipt and bait the fake Bon Jovi, giving the victim some much-needed closure.

Scamming a romance scammer

Broadcasting on National Geographic at 9 pm ET and available to stream on Disney+ and Hulu, the documentary takes viewers on an empathetic quest to understand how the line between “lover” and “fraud victim” gets blurred in romance scams that the Federal Trade Commission says cost Americans nearly $1.5 billion in losses each year.

Read full article

Comments

© National Geographic