US senators propose curbs on AI-generated election deception

A group of Senate Democrats Thursday renewed a push to regulate the use of AI in federal elections, targeting both synthetic media designed to mislead voters and automated or database-driven voter challenges that voting rights advocates warn could be used to remove eligible voters from the rolls.

The Fraudulent Artificial Intelligence Regulations Elections Act (FAIR), introduced by Sens. Jeff Merkley and Alex Padilla, would prohibit the knowing distribution of false AI-generated election content intended to stop people from voting.

It would also amend the National Voter Registration Act to restrict the removal of voters from registration lists unless state officials rely on adequate and verified information.

The bill is aimed at a convergence of two election threats that have grown more urgent since 2020, the use of AI to generate highly scalable election disinformation, and the use of scraped, mismatched, or otherwise unreliable data to mount mass voter challenges.

The bill’s central premise is that AI can be used to confuse voters about when, where, and how to vote, impersonate trusted officials, or accelerate challenge operations that overwhelm election administrators and put lawful registrations at risk.

In February, Biometric Update reported that AI-generated deepfake videos and photographs were expected to be deployed not only to mislead voters in the run-up to the mid-term elections, but to intimidate them into not voting.

Merkley said the bill he introduced this week is intended to stop bad actors from using AI “to spread disinformation about voting” as part of voter suppression efforts.

Padilla, California’s former secretary of state, framed the threat as both technological and historical, saying voters should not be blocked from participating because they are misled about how to vote or because unverified databases force them off the rolls.

The bill comes as federal and state policymakers are still trying to catch up with generative AI’s role in political speech, election administration, and disinformation.

In the 2024 cycle, election officials and regulators confronted one of the clearest early examples of the threat when robocalls using an AI-generated voice mimicking President Joe Biden were sent to New Hampshire voters before the state’s presidential primary.

The calls falsely suggested that voters should stay home and save their vote for November. New Hampshire officials traced the calls to Life Corporation and Lingo Telecom. The Federal Communications Commission later reached a $1 million settlement with Lingo over the transmission of the calls.

The FAIR Act would make that kind of conduct easier for federal authorities to stop when it is tied to federal elections.

The bill defines “false AI-generated election media” as text, images, audio, or video that is created through machine learning, natural language processing, AI techniques, or similarly complex computational processes, when the content either contains materially false information about an election or falsely depicts an election official.

It would apply to false AI-generated content about the time, place, or manner of holding a federal election, or about voter eligibility requirements and restrictions when the person distributing it knows the information is materially false and acts with the purpose of impeding or preventing another person from voting.

It would also cover false AI-generated depictions of election officials when the distributor knows the depiction is false and intends to intimidate or harass the official or deter another person from voting.

Enforcement would rest with the attorney general, who could bring a civil action in federal court seeking declaratory and injunctive relief, including temporary restraining orders and permanent or temporary injunctions.

The bill does not create a general criminal offense for AI-generated election disinformation. Instead, it gives the Justice Department a civil tool to stop or restrain targeted false AI election media when it is knowingly distributed for voter suppression or election-worker intimidation.

The second major portion of the legislation addresses voter-roll removals and mass challenges.

The bill would amend the National Voter Registration Act by adding new restrictions on when a state may remove a registrant from the official list of voters eligible to vote in federal elections or act on a voter eligibility challenge.

A state could not do so unless the voter is determined to be ineligible based on source information received from governmental entities in the course of official duties, and source information approved by the attorney general as sufficient to protect the integrity and completeness of voter registration lists.

That provision is aimed squarely at the rise of privately generated voter challenge databases and software tools that voting rights organizations say rely on scraped public records, commercial data, outdated lists, weak matching logic or other error-prone methods.

Ever since the 2020 election, largely MAGA aligned groups claiming to search for voter fraud have used tools such as EagleAI, VoteRef, Check My Vote, and IV3 to generate large numbers of voter challenges.

Under the bill, source information approved by the attorney general would have to contain qualifying data sets that allow a state to match the individual identified by the source information with an individual on the voter list. It would have to be updated at least monthly. And it could not have been obtained in violation of the Computer Fraud and Abuse Act.

The bill defines qualifying data sets to include combinations such as last name, first name and full Social Security number; last name, first name, and a driver’s license or other unique state identifying number; last name, first name, the last four digits of a Social Security number and date of birth; or last name, full Social Security number, and date of birth.

Those requirements would make it harder for a voter to be removed based only on loose name matches, commercial data, scraped web records, or automated challenge lists that lack strong identifiers.

It would prohibit the federal government from comparing personal information held by the federal government against state or local government data for the purpose of determining eligibility to vote in federal elections.

It would also create a private right of action allowing an aggrieved person to sue the federal government for declaratory or injunctive relief if the prohibition is violated.

That provision is linked to growing Democratic concern over the use of federal databases, including the Department of Homeland Security’s Systematic Alien Verification for Entitlements system in voter eligibility checks.

Supporters of the legislation argue that AI changes the scale and speed of old voter suppression tactics. False information about polling places, voting dates, or eligibility has long been used to confuse voters, while mass challenges have long been used to burden election offices and place voters on the defensive.

Article Topics

AI fraud  |  deepfakes  |  elections  |  generative AI  |  legislation  |  U.S. Government  |  voter registration