Fake videos, edited calls and recycled rumours spread faster than platforms can remove them
AI disinformation is flooding social media before the US midterms, and experts say moderation lags, TF reporting finds.
Voters go to the polls in November. The platforms carrying the claims have cut staff who once policed them.
What’s Happening & Why It Matters
Cheap Fakes, Endless Versions
Frances Haugen, the former Facebook data engineer and whistleblower, says AI makes influence operations cheap and convincing. Operators can produce many versions of one message. Small edits let a post slip past filters after a removal. She argues platforms should study how content spreads and which feedback loops slow spread, instead of chasing single posts.
Trust and safety teams have shrunk since a backlash from the right over speech limits. Platforms lean on AI moderation and community tools, and researchers doubt either catches enough.
The Texas and Ohio Fakes
In Texas, a false story said Democratic Senate candidate James Talarico was in an Ohio mental health facility. The rumour grew from a doctored phone call. His campaign published security-camera footage placing him at home in Texas, and he said he had never been to Ohio. The Digital Democracy Institute of the Americas (DDIA) found the story mutated into a suicide-attempt claim in Spanish-language conservative media. Roberta Braga, its founder and executive director, says platforms enforce rules with less rigour in Spanish than in English.
Talarico’s Republican opponent, Ken Paxton, ran an ad with AI-generated footage of Talarico rewriting the Bible. AP reports other videos targeting Talarico and Ohio’s Sherrod Brown. Clemson researcher Darren Linvill links them to Storm-1516, a Russian network active in 2024. Russia’s embassy did not reply to AP. X removed the first posts and suspended the accounts, but copies stayed up.
The Pennsylvania Claim
Viral posts claimed an undocumented immigrant arrested in central Pennsylvania worked for Democrat Janelle Stelson. She challenges Republican Scott Perry. Her campaign released records showing the man held no paid role. A police report says he claimed to be a worker.
A statement from the Department of Homeland Security amplified the claim. Perry’s campaign still runs ads built on the story. The watchdog group CREW filed a Hatch Act complaint against the department over the statement. Free Press attorney Nora Benavidez says federal agencies have been used in ways which interfere with the elections.
What the Platforms Promise
Meta (@Meta) requires AI disclosure in election ads and labels AI images using C2PA data. It bans false voting dates and places, yet does little to police political commentary. YouTube bans false voting logistics, and stopped removing claims of fraud in past elections in 2023. X relies on Community Notes, and experts say enforcement is weak.
My take: Platforms are weakest where damage is hardest to undo: the final week before a vote and non-English feeds. Rules on paper do not help a voter who sees a fake once and never sees the correction. Campaigns share the blame when their ads use synthetic footage. Readers should check a clip against the candidate’s official channel before sharing.
TF Summary: What’s Next
Voting takes place on 3 November 2026. Expect more fakes targeting close races and Spanish-speaking voters. Campaigns will keep issuing takedown demands, and lawyers will test platform and agency conduct after the count.
MY FORECAST: A fabricated clip will gain over a million views in a swing race before election day. At least one campaign will sue a platform or media outlet over synthetic content by January 2027. Congress will hold a hearing on AI election fakes in the next session.
Related Stories
- Norway Expands Under-16 Social Media Ban to Games and Messaging Apps
- Meta Faces $25 Billion Fine in New Mexico Privacy Case
- Claude Bypasses Georgia Paywall as Elvis Estate Sues Anthropic
- US Bars DeepSeek From Federal Civilian Agencies
- Sweden Blocks ChatGPT and Claude on Government Computers