Voting Rights Groups Warn About AI Generating Unfounded Claims in Spanish

Two voting rights groups have issued warnings this month about the possibility of artificial intelligence (AI) generating unfounded claims about the upcoming election in Spanish.
With just days until the presidential election, Latino voters are seeing a surge of Spanish-language ads and a new wave of political content shaped by artificial intelligence: Chatbots spreading unverified claims in Spanish about voting rights.
AI models are generating election-related misinformation in Spanish at a higher rate than in English, complicating the flow of reliable information for one of America’s fastest-growing and increasingly influential voting blocs, according to an analysis by two nonprofit news organizations.
Voting rights advocates warn that AI-driven misinformation could widen information gaps for Spanish-speaking voters, a crucial demographic targeted by both Democrats and Republicans in races across the ballot.
Proof News and Factchequeado, in partnership with the Science, Technology and Social Values Lab at the Institute for Advanced Study, evaluated popular AI models’ responses to targeted prompts ahead of the November 5 election, assessing the accuracy of their outputs.
The study found that over half of the election-related responses generated in Spanish contained inaccuracies, compared to 43 percent of responses in English.
Voting rights advocates have warned for months about a surge of online and AI-driven misinformation targeting Spanish-speaking voters. The new analysis underscores the importance of verifying election information, said Lydia Guzman, who heads a voter advocacy initiative at Chicanos Por La Causa.
“It’s important for every voter to do proper research and not just at one entity, at several, to see together the right information and ask credible organizations for the right information,” Guzman said to The Associated Press (AP).
Large language models, trained on extensive datasets sourced from the internet, can generate AI-driven responses but remain prone to inaccuracies. Even if Spanish-speaking voters aren’t directly interacting with chatbots, they may still encounter AI-generated content embedded in tools, apps or websites that rely on these models.
These inaccuracies could carry significant impact in states with large Hispanic populations, including Arizona, Nevada, Florida and California.
In California, nearly one-third of all eligible voters are Latino, and for one in five Latino voters, Spanish is their only language, according to the UCLA Latino Policy and Politics Institute.
Meta’s AI model Llama 3, which powers the virtual assistant in WhatsApp and Facebook Messenger, performed poorly in the test with nearly two-thirds of its Spanish responses containing inaccuracies, compared to about half of its English responses.
For instance, Meta’s AI misinterpreted a question about “federal only” voters. In Arizona, these voters are those who registered without providing state-required proof of citizenship—typically because they used a federal form that doesn’t ask for it—and are eligible only for presidential and congressional races. Instead, Meta’s AI incorrectly stated that “federal only” voters are U.S. territory residents, such as those in Puerto Rico or Guam, who actually cannot vote in presidential elections.
Meta spokesperson Tracy Clayton told the AP that Llama 3 is designed as a tool for developers to create other products. He added that Meta is training its models with safety and responsibility guidelines to reduce the risk of inaccurate voting information being shared.
This article includes reporting from The Associated Press.