Connect with us

NEWS

Campaigns Have No One to Call as AI Lies Spread

Ahead of the November 3 midterms, AI lies move faster than platform takedowns, and campaigns now face federal agencies repeating the same viral fabrications.

Published

on

Social media companies have left US campaigns with no desk to call as AI-made election lies spread ahead of the November 3 midterms. Trust and safety teams that once fielded those calls were cut after a years-long fight over speech rules. The people running Senate and House races now chase fabrications that mutate overnight, and in one Pennsylvania contest a federal agency repeated the claim in a press release.

The cheap part is new. The empty phone line is what changed the job.

Nobody Left for Campaigns to Call

Frances Haugen, a former Facebook data engineer, said a dirty-tricks shop can now run an information operation “with quality that was impossible in 2016 for super, super cheap.” Operators keep altering a post until a slice of the copies clear the filters. Some of them, she said, are going to get through.

That is the bind for campaigns from Texas to Pennsylvania. Meta, which owns Facebook and Instagram, and X did not answer questions about the current wave. YouTube, owned by Google, said every user faces the same community guidelines and the same public reporting tools. Those tools were built for a slower internet. They were not built for a candidate who needs a human being to pull a lie before it is translated, recut, and paid out as an ad.

Haugen’s next point is the one campaigns are living. When the model can tweak a clip until it slips past a classifier, takedowns aimed at a single piece of content stop being a plan. The fight moves to distribution, feedback loops, and who still answers the phone.

How a Flu Absence Became a Suicide Claim

James Talarico, the Democratic Senate candidate in Texas and a Presbyterian seminarian, left the trail in early October with the flu. He had already said so at a September 30 Hispanic Business Council forum he joined by video. Conservative influencers Nick Sortor and Rhein Amacher then posted that they had called the Lindner Center of Hope, a mental health hospital in Mason, Ohio, and been told he was a patient. Sortor put up a 47-second clip of the call. Ken Paxton, the Republican nominee and still Texas attorney general, posted that Talarico was having a “full-blown mental Tala-freak-out” and asked for “proof of life.”

The hospital knocked the story down on October 11. Talarico quoted the denial and added a line of his own.

Home security images covering October 1 through October 8 put him at his Austin house. He was at the Austin City Limits festival on October 9 and at a Kacey Musgraves show in Dallas on October 10, the window in which the clip claimed he had been discharged. The full call, later obtained in unedited form, shows a staffer denying any record of him. JT Ennis, a Talarico spokesman, called Sortor’s post “a lie designed to promote a pathetic conspiracy theory.”

THE OHIO HOAX, BY DATE

  1. September 30, 2026: Talarico joins a forum by video and says he has the flu.
  2. October 1 through October 8, 2026: Security-camera images show him at home in Austin. He makes no public appearances.
  3. October 9, 2026: He appears at Austin City Limits.
  4. October 10, 2026: He is seen at a concert in Dallas.
  5. October 11, 2026: Influencers post the edited Ohio call. Lindner Center of Hope says he “has never received treatment” there. Talarico replies, “I’ve never even been to Ohio.”

The Digital Democracy Institute of the Americas tracked what happened next in Spanish-language channels. The English lie (hospitalized in Ohio) became a false claim that he had tried to kill himself. DDIA researchers wrote that lies evolve as they travel, “making the original correction insufficient to address every version of the story.” An English-language punchline does not retire a Spanish-language suicide rumor.

Paxton had already tested synthetic video on Talarico. A campaign ad earlier in the year used AI-generated footage of the seminarian supposedly rewriting the Bible. “These AI deepfake ads don’t add anything to our civic discourse,” Talarico said. “They don’t do anything to elucidate the choice that’s in front of voters.”

Late-cycle clips kept coming. Hany Farid, a Dartmouth College professor who studies deepfakes, said a video framed as a former student’s confession about Talarico was likely generated or altered with AI and fit the pattern of smears released close to Election Day. Ennis said the people who cannot beat Talarico “with the truth” were “slandering him with AI deepfakes, foreign bots, and conspiracy theories.”

An ICE Statement Named the Democrat

In Pennsylvania’s 10th District, the lie did not stay on blogs. Janelle Stelson, the Democratic nominee challenging Republican Rep. Scott Perry, spent October trying to catch a story that an undocumented immigrant arrested near Harrisburg was her canvasser.

East Pennsboro Township police arrested Chimaobim Adedayo Igbojekwe, a Nigerian national, on September 27 after an officer wrote that his genitals were exposed in a parked car. The report says Igbojekwe told the officer he was waiting to hand out flyers for Stelson and produced a stack of them. He later pleaded guilty to disorderly conduct. The indecent exposure count was withdrawn. ICE arrested him on October 1.

On October 7, the Department of Homeland Security issued a release that treated the flyer story as employment. Charles Wall, ICE’s deputy director, said it was “unbelievable that Janelle Stelson would employ an illegal alien to work on a congressional campaign, let alone a pervert who exposed himself in public.”

Stelson answered the next day. “He’s not an employee. He’s not a volunteer.” Alma Baker, her campaign manager, said the campaign had never employed him, that his name is absent from FEC disbursements and the volunteer list, and that he had not worked for campaign vendors. Baker said the campaign was “likely to take legal action to combat the blatant lies and coordinated misinformation campaign.” Perry’s operation kept running the claim in ads and texts. The Stelson team sent cease-and-desist letters. A television reporter deleted a post about the arrest. The campaign still had not filed a complaint with Meta, X, or YouTube, and a spokesperson said it was weighing “every legal avenue.”

WHAT WE KNOW

  • The arrest: Igbojekwe was arrested locally on September 27 and by ICE on October 1, then pleaded guilty to disorderly conduct.
  • The flyers: Police wrote that he had Stelson literature and said he was there to hand it out.
  • The campaign books: Stelson’s team says he was never an employee, volunteer, or vendor hire, and FEC records show no payments to him.
  • The federal line: DHS named Stelson in an October 7 release, and CREW asked the Office of Special Counsel to treat Wall’s quote as a Hatch Act problem. OSC said it would open a file.

WHAT IS UNCONFIRMED

  • Who supplied the flyers: Police described a stack in the car. The campaign says it never heard of him. It is not established who printed or handed him the literature.
  • A paid outside canvasser: It remains unclear whether some other group, not the campaign, had him in the field.
  • The Hatch Act outcome: OSC opened a file. It has not issued a finding.

Citizens for Responsibility and Ethics in Washington argued that naming a candidate and attacking her over a canvasser, weeks before a vote, had “no other purpose than to cast aspersions on candidate Stelson.” Donald Sherman, CREW’s executive director, called Wall’s line “gratuitous partisanship that constitutes prohibited political activity.” Federal employees may not use using official authority to affect an election as a political tool under 5 U.S.C. § 7323. Nora Benavidez, a civil rights lawyer with Free Press, said the White House has pulled agencies from the Postal Service to DHS into the fight.

Online hit squads used to live on the fringe. An ICE deputy putting the same story on letterhead is a different distribution system.

What the Platforms Still Take Down

The large platforms do have rules. They are narrow. They are built around how to vote, not around a fake hospital stay or a disputed canvasser, and they leave most candidate smears in place if the clip never mentions a polling place.

WHERE THE RULES STOP

Platform Still banned Left standing
Meta (Facebook, Instagram) Ads and posts that misstate voting dates, locations, or legal requirements; threats of violence; undisclosed synthetic election ads when detected Most political commentary, including debate over whether past elections were legitimate; unsolicited political posts kept out of some recommendation slots rather than deleted
YouTube Content that misleads about the time, place, means, or eligibility to vote; false claims meant to discourage voting; calls to hack polling computers or incite violence at voting sites False claims of widespread fraud in 2020 and earlier presidential races, under the June 2, 2023 policy; a direct lie about a candidate that never touches voting mechanics
X False information about how to vote and suppression schemes; synthetic media “designed to mislead,” on paper Most civic claims handled through Community Notes after the fact, with no internal safety review of the kind the company used before Elon Musk’s purchase

Meta requires political advertisers to disclose synthetic images and AI-made audio or video, and it applies automated labels when content provenance metadata from the Coalition for Content Provenance and Authenticity is present. It also bans posts that misrepresent when, where, or how to vote. It does little active policing of political argument beyond keeping some unsolicited political content out of recommended slots. Free Press has noted that Meta narrowed older limits in 2023 so users could again argue over the legitimacy of past cycles.

On February 19, 2026, Meta set out its midterm plan: a publicly searchable Ad Library that already holds more than 18 million US entries, kept for seven years; voting alerts that state and local officials have used to send more than 1 billion notifications; and Community Notes on Facebook, Instagram, and Threads, which publish only when contributors who usually disagree reach consensus. The company says it has spent more than $30 billion on safety and security over the last decade and has removed 200 coordinated inauthentic-behavior networks since 2017. It also announced a final-week political ads restriction, blocking new political, electoral, and social-issue ads in the last week of the campaign. On October 27 it said that restriction period had started at 12:00 a.m. PT and that its Election Operations Center was active.

Google would rather raise official pages than delete a nasty clip. Civic leaders can ask YouTube to take down a video that uses an AI-generated likeness to fake their words or deeds. A lie that never trips those lines can sit. X’s civic integrity policy still bars false how-to-vote information. Experts say a synthetic-media rule with no enforcement staff is a paragraph, not a brake.

YouTube’s 2023 Rollback Still Governs Feeds

The empty desk is not an accident of this month. It is the policy the companies wrote in 2023 and then staffed down.

After Cambridge Analytica, the large platforms hired trust and safety workers in bulk. The counter-swing came from the right, over speech rules, and hardened after Musk bought Twitter and Donald Trump won in 2024. Those teams were then cut. The companies leaned on automated systems to do the work humans used to do. Haugen’s warning sits on top of that choice: the same class of model that writes the lie is the class of model that is asked to catch it.

When you stop being able to focus on content because the AI will just slightly tweak it until it gets through, you have to focus on what are the systems of distribution.

Frances Haugen, former Facebook data engineer

YouTube made the underlying speech choice explicit on June 2, 2023. After tens of thousands of removals, it said it would stop removing past election fraud claims that the 2020 race and earlier presidential elections were marred by “widespread fraud, errors, or glitches.” Removals, the company wrote, did curb some falsehoods but also risked “curtailing political speech without meaningfully reducing the risk of violence or other real-world harm.” Time, place, means, and eligibility lies stayed banned. So did claims meant to talk people out of voting. The company said 2020 election information panels had been shown more than 4.5 billion times and that it would keep putting official sources high in search and recommendations.

That is the machine Talarico and Stelson are feeding a correction into. A Community Note, an information panel, and a campaign statement all arrive after the first version has already been copied, translated, and clipped.

Spanish-Language Posts Face Thinner Review

Roberta Braga, founder and executive director of DDIA, said the companies enforce their rules “more congruently” in English than in Spanish. “There’s a lot less guardrails in place from the companies to deal with Spanish language, especially now that they’ve walked back a lot of their own investments in content moderation, and they’ve moved toward doing community-centered content moderation, like Community Notes-style things,” she said. “Now that they’re using AI to do a lot of the content moderation, even more is likely slipping through the cracks.”

DDIA’s July snapshot of Latino spaces, covering January through May, treated that gap as predictable. The large false story barely changes (elections cannot be trusted). The supporting claims rotate. The institute listed 10 recycled election-fraud claims it expected to see again before November 3, from non-citizens voting to dead voters, jammed machines, and invented quotes hung on candidates.

THE FOUR BUCKETS DDIA TRACKS

  • The system: Hacks, switched ballots, and “deep state” rigging, including Venezuela-themed interference claims.
  • The process: Lies about who may vote, what ID to bring, and what a delay at a precinct is supposed to prove.
  • The voters: Stories about non-citizens or the dead, often built from a single real case that could not change a result.
  • The candidates: Fabricated statements and comments stripped of context, the bucket that caught Talarico.

In those five months DDIA logged almost 300 unique Spanish-language WhatsApp messages about election integrity in public groups, about two a day, with a potential reach of more than 20,000 users in data from Palver. A recycled transnational story still tied US voting machines to Venezuelan politics, naming Dominion and Smartmatic, a claim US officials, courts, and cybersecurity agencies have already rejected. On X, Braga’s “breaking news” layer moves first. YouTube then hosts long Spanish videos that stitch several of those claims into one plot. TikTok supplies the short outrage clip.

Paxton had already used the cheaper toolkit on a fellow Republican. In mid-January he shared an AI-generated video of then-primary rival Sen. John Cornyn dancing with Democratic Rep. Jasmine Crockett. Officials with huge followings have also pushed grotesque synthetic clips of rivals and treated the blowback as comedy. That lowers the bar for the next fake, including the ones that never carry a label.

$137 Million in AI Ads, Mostly Unlabeled

Organic lies are only half the pipe. Paid ones are the half with a rate card.

The Wesleyan Media Project, a research group at Wesleyan University that tracks election advertising with AdImpact, identified 220 political ads generated or enhanced with AI across 34 states through October 11. Sponsors spent more than $137 million on them. In the 2024 cycle the same researchers found less than $50,000 in AI political ads. Republican candidates and allied groups sponsored 81 percent of the 2026 spots; Democrats sponsored about 12 percent. Only 27 percent carried a disclaimer that AI was used. Erika Franklin Fowler, the project’s co-director, called the spread “problematic” and said the tally was almost certainly low, because use is hard to spot when no one labels it.

THE PAID LAYER

  • The spend: More than $137 million on AI-made or AI-enhanced political ads through October 11, per the Wesleyan Media Project.
  • The volume: 220 spots in 34 states, against less than $50,000 in such ads in 2024.
  • The label rate: 27 percent of those ads disclosed AI use.
  • The split: 81 percent of the ads came from Republican sponsors, about 12 percent from Democrats.

Meta’s own rules still ask advertisers to disclose when a social-issue or election ad uses third-party generative tools to make a real person say or do something they did not, to invent a realistic person or event, or to alter footage of a real event. From June 1, 2026, the company said it would also auto-detect some of that material and attach an “AI Info” label in an “About this ad” panel. Ads already running before the late-October restriction can keep running. A smear that launched in September does not need a new insertion order in the final week.

That is the second-order bind on November 3. A candidate can post security-camera stills, a hospital can deny a patient record, and a campaign manager can wave FEC files, and none of those moves summons a room of reviewers. The same week, a federal law-enforcement statement can give the first version a second life. The platforms will still pull a post that names the wrong Election Day. They will not, as a matter of written policy, pull a deepfake of a seminarian rewriting scripture, a suicide rumor that only exists in Spanish, or a canvasser story that an ICE deputy already retold. The vote is on November 3. The phone on the other end of the old trust-and-safety line does not ring.

Harry is the editor and publisher of MY WORLD NEWS 24, an independent title under his own ownership. Ten years of reporting and then editing taught him that a global readership is not served by assuming everyone lives in the same country. Stories here state currencies, units and time zones explicitly, name the country a law or a company belongs to, and explain local context rather than treating it as known. That care extends to sourcing: a claim is anchored to the filing, statement, transcript or dataset that made it, wherever in the world it was issued, and each figure is checked against that source before publication. The site reports news, business and technology, science and sports, entertainment, lifestyle and travel, and auto and gaming, all with the same standard of evidence. Mistakes are fixed under a corrections policy anyone can read, and the page carries a note saying what was changed. Harry reads every message sent by readers and replies from support@myworldnews24.com.

Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Trending