The United States Department of Homeland Security (DHS) used artificial intelligence (AI) tools and commercial consumer data to identify people suspected of unlawful voting, according to court documents, prompting a federal judge to restrict election-related investigations in nine states ahead of the November midterm elections.
US District Judge Araceli Martínez-Olguín issued a preliminary injunction on 21 October, finding that the government's voter-targeting initiative relied on unreliable data and risked intimidating eligible voters.
The ruling limits field operations against people agents know or reasonably suspect are US citizens, while allowing investigations of suspected non-citizens to continue within the order's restrictions.
AI Tools Processed Millions of Voter Records
The initiative, known as the Unlawful Voter Initiative, was introduced in August 2026 to identify people suspected of registering or voting unlawfully.
DHS directed US Citizenship and Immigration Services (USCIS) personnel to compile target lists and coordinate with Immigration and Customs Enforcement (ICE) investigators.
Court documents show that DHS received 48.2 million records from independent sources of questionable provenance, alongside 50.6 million records from official state voter rolls or commercial versions of those records.
The department used a software system called iLab, incorporating unspecified AI tools, to combine information from different sources and create individual profiles.
The agency also used commercial databases typically associated with targeted advertising. These contained personal information including residential addresses, geolocation data, demographics, family relationships, vehicle ownership and consumer behaviour.
Officials used these records to review and eliminate individuals from the target lists, although the court noted that the original information was unreliable.
Officials Had 12 Minutes to Review Suspected Voters
The court found that the initiative's screening process carried a significant risk of mistaken identification. USCIS acknowledged that similar names, incomplete records, outdated citizenship information and clerical errors could lead to US citizens being flagged for further investigation.
During an intensified phase of the initiative, personnel were instructed to review at least 40 suspected individuals daily. That requirement left investigators no more than 12 minutes to eliminate each person from the target list using commercial databases, according to the court documents.
The order also highlighted problems with previous voter-data checks. DHS initially identified 15,903 potential non-citizens on Nevada's voter rolls, but the figure fell to 185 after state officials questioned the results and eventually dropped to zero.
The case documents further state that the initiative targeted naturalised citizens when records suggested they had voted before obtaining citizenship.
Judge Restricts Field Operations in Nine States
The injunction covers California, Connecticut, Georgia, Missouri, Nevada, New York, Pennsylvania, Washington and Wisconsin.
The judge found that the investigations could create a reasonable fear of government scrutiny among eligible voters and people helping others participate in elections.
The lawsuit was brought by the California Nurses Association, National Nurses United, the city and county of San Francisco, Santa Clara County and Voto Latino Foundation. The groups challenged the initiative on the grounds that the investigations could violate federal protections against voter intimidation.
The order took effect after a temporary pause to allow the government time to appeal and is set to expire after Election Day on 3 November.
It does not prohibit all investigations into suspected non-citizen voting but restricts field operations targeting people known or reasonably suspected to be US citizens.
