Georgia holds emergency meeting on AI exposing voters’ secret ballots

Georgia holds emergency meeting on AI exposing voters’ secret ballots A Princeton researcher found that publicly available election records could be combined with AI to link voters to their ballots

When voters cast their ballots, their votes are supposed to remain secret: from their family, their neighbors, and the government.

But what if artificial intelligence could make secret votes visible?

“Within a couple of hours, I had a pipeline to analyze and identify secret ballots across the state of Georgia and the agent told me exactly what further information it would need to identify real voters’ ballots,” Springer wrote. “At no point did the agent refuse to comply or raise concerns over implementing the exploit.”

The issue has raised alarms among Georgia’s election officials less than two weeks before early voting starts in the high-stakes midterm election that will determine control of the US Congress.

Ballot secrecy is a democratic safeguard , allowing voters to shield their selection from anyone who might try to influence or intimidate them into voting a certain way. But Springer’s test showed that he could use AI to find out who people voted for, using Georgia’s public election records.

Georgia’s state elections board held an emergency meeting on Thursday morning to discuss how artificial intelligence has made it easier to identify voters through their ballot code, leaving people more vulnerable.

Georgia’s outgoing secretary of state, Brad Raffensperger, locked down the tabulation data that someone might use to identify a voter, ordering any public release to redact the ID numbers after an election, which “substantially address[es] this flaw”, said Ben Adida, an MIT-trained cryptographer and the founder of the election technology non-profit VotingWorks, as he addressed the board on Thursday.

Election workers make a record of every voter who comes through a polling place. Voters then choose candidates on a touchscreen, which then prints out a paper ballot. The voter feeds that ballot by hand into a scanner to be tabulated.

“When a voter inserts their ballot into a tabulator, a digital record of that ballot is created. Think of it as a row in a spreadsheet. One row is one scanned ballot. It’s often called a cast vote record, or CVR for short. In order to later retrieve the CVR, the tabulator assigns a unique identifier to each CVR at the time of scanning.”

The flaw shows up in how the machine records that ballot, Adida said. The identifying is not actually random enough.

Using AI, the early-voting list and the cast-vote record file for each county, Springer, the Princeton researcher who says he’s never been to Georgia, could recover the voting order for 1.52m ballots, or 98.9% of in-person ballots in 114 of the 139 counties he examined.