48 hours: The window in which AI election lies outpace corrections

Social media platforms are failing to contain the spread of AI-generated disinformation ahead of the November midterm elections, a gap that has left campaigns scrambling to correct falsehoods that evolve faster than human moderators can react. As tens of millions of Americans prepare to vote, major companies have reduced their trust and safety staffs, relying instead on artificial intelligence to detect and remove deceptive content. This shift has created a vulnerability where minor variations of a lie can bypass automated filters, allowing misinformation to settle into the public record before a candidate can respond.

The timeline of these efforts stretches back to the backlash following the Cambridge Analytica scandal, which initially drove Facebook and other giants to expand their moderation teams. However, a counter-reaction from the right, intensified by Elon Musk's acquisition of X and the 2024 election of Donald Trump, led to significant layoffs in these departments. Frances Haugen, a former Facebook data engineer and whistleblower, noted that the current environment allows operatives to run high-quality information operations at a fraction of the cost of 2016. She explained that creators can tweak posts slightly to evade detection, causing a high volume of false content to slip through the cracks of AI-driven moderation.

The consequences are visible in specific races. In Texas, Democratic Senate candidate James Talarico took time off campaigning earlier this month to recover from the flu. During this absence, a fabricated story spread across social feeds claiming he was being treated at a mental health facility in Ohio. The claim originated from a right-wing provocateur who published a deceptively edited phone call with a hospital staffer. Talarico’s campaign released security camera footage showing him at home in Texas and stated he had never been to Ohio, but the lie persisted for several days.

The Digital Democracy Institute of the Americas tracked how the Talarico story mutated as it spread. The initial viral claim in English about hospitalization in Ohio transformed into a false assertion in Spanish that the candidate had attempted suicide. Roberta Braga, the institute’s founder, pointed out that platforms enforce terms of service more rigorously in English than in other languages. She noted that with reduced investments in human moderation and a move toward community-centered tools like Community Notes, AI-driven moderation is missing more errors.

In Pennsylvania, Democratic congressional candidate Janelle Stelson faced a similar challenge when viral posts claimed an undocumented immigrant arrested in central Pennsylvania was an active field worker for her campaign. While the campaign released records proving the man, Chimaobim Adedayo Igbojekwe, had no paid position, the claim was amplified by an official statement from the Department of Homeland Security. Charles Wall, the deputy director of Immigration and Customs Enforcement, declared it unbelievable that Stelson would employ an illegal alien and a pervert. The opposing Republican campaign, led by Congressman Scott Perry, continues to run ads using the claim.

Platform rules regarding AI election misinformation are narrow and often passive. Meta requires advertisers to disclose synthetic media in election ads and bans posts that misrepresent voting dates or locations. However, the company does not actively moderate political commentary beyond excluding unsolicited content from recommendations. YouTube stopped removing content that falsely claims past elections were marred by widespread fraud in 2023, citing a chilling effect on speech. The platform now relies on users to report violations and uses contextual information panels to amplify authoritative sources rather than removing content.

X, formerly Twitter, abandoned internal safety reviews after its acquisition by Musk, relying on the crowdsourced community notes program. Its policy bars false information about voting mechanics and suppression schemes but restricts synthetic media only when designed to mislead. Experts argue this policy lacks teeth without active enforcement. Haugen suggested that when content is constantly tweaked to evade filters, the focus must shift to the systems of distribution and the biases within feedback loops that allow misinformation to dampen or amplify.

Campaigns are currently using legal avenues to fight these claims. Stelson’s team sent cease-and-desist letters to media outlets and activists filed a complaint accusing DHS of violating the Hatch Act, which limits political activities of government employees. Talarico told NBC News in September that Texans are fed up with AI deepfake ads that do not elucidate the choices in front of voters. The next step for these campaigns involves continuing to monitor platform enforcement while preparing for the final weeks of the election cycle.