AI disinformation surges as US platforms cut human review before midterms

Social media platforms are facing a surge of artificial intelligence-driven false information ahead of the United States midterm elections in November. Media experts state that these fabricated posts are evading moderation systems just as tens of millions of Americans prepare to cast their ballots. The increase in automated content creation has outpaced the ability of human reviewers to verify accuracy, leading to a higher volume of errors reaching voters.

The shift in platform strategy stems from a political counter-backlash following the 2024 election. After the Cambridge Analytica scandal, major tech companies expanded their trust and safety teams. However, following Elon Musk's acquisition of X and Donald Trump's victory, concerns over speech restrictions grew. This led to significant layoffs in moderation departments and a greater reliance on AI tools to police user content.

Frances Haugen, a former Facebook data engineer and whistleblower, noted that AI has altered the cost of running information operations. She stated, "You can run an information operation now with quality that was impossible in 2016 for super, super cheap." Haugen explained that operatives can continuously adjust posts until they pass moderation filters, meaning "some of them are going to get through."

In Texas, Democratic Senate candidate James Talarico recovered from the flu earlier this month. During his absence, a false story circulated on social media claiming he was being treated at a mental health facility in Ohio. A right-wing figure promoted the claim using a deceptively edited phone call with a hospital employee. Talarico's campaign released security footage showing him at home in Texas, and he confirmed he had never visited Ohio. The false narrative persisted for several days before subsiding.

The Digital Democracy Institute of the Americas tracked how the story mutated in conservative Spanish-language media, which later claimed Talarico attempted suicide. Roberta Braga, the institute's executive director, said platforms enforce rules more consistently in English than in Spanish. She added that increased use of AI moderation means "even more is likely slipping through the cracks." Talarico's Republican opponent, Ken Paxton, also ran an ad earlier this year featuring AI-generated footage of Talarico rewriting the Bible.

In Pennsylvania, viral posts claimed that Chimaobim Adedayo Igbojekwe, an undocumented immigrant arrested in the state, was a canvasser for Democrat Janelle Stelson's congressional campaign. ICE deputy director Charles Wall amplified the claim, stating it was unbelievable that Stelson would employ an illegal alien. Stelson's campaign provided records showing Igbojekwe held no paid position, though an initial police report noted he claimed to be a campaign worker. Republican congressman Scott Perry's campaign continues to run ads containing the claim.

Stelson's team sent cease-and-desist letters to media outlets, and a Fox News reporter removed a related post on X. Citizens for Responsibility and Ethics in Washington filed a complaint alleging a Hatch Act violation by the Department of Homeland Security. Nora Benavidez of the Free Press said Trump "has made clear that he intends to interfere in the elections" using federal agencies. Stelson's spokesperson said the campaign is considering every legal avenue against the false claims.

Major platforms maintain specific rules for election content. Meta requires disclosure of AI content in election ads and labels AI images identified through C2PA metadata. It bans posts that misstate voting dates or locations but does little to moderate political commentary. YouTube stopped removing false claims about past elections in 2023 and defines misinformation narrowly. X replaced internal safety reviews with crowdsourced community notes, which experts say lacks consistent enforcement.

Haugen argued that moderating individual posts is insufficient when AI can tweak content to bypass filters. She said attention must turn to the systems of distribution, including their biases and feedback cycles that allow communities to dampen harmful content.