Georgia addresses AI vulnerability exposing secret ballots after researcher's test

AI Exposes Ballot Secrecy

Max Springer, a postdoctoral research fellow at Princeton University, demonstrated that artificial intelligence could be used to identify voters based on their Georgia ballots. Using a $20 AI large language model subscription and data obtained through an Open Records Act request, Springer developed a method to analyze and identify secret ballots across the state.

Emergency Response from Georgia Officials

This discovery prompted Georgia’s state elections board to hold an emergency meeting. The concern centered on how AI made it easier to link voters to their ballot choices, potentially compromising the democratic safeguard of ballot secrecy, which protects voters from influence or intimidation.

Mitigation Measures Implemented

In response, Georgia’s outgoing Secretary of State, Brad Raffensperger, ordered that any public release of tabulation data must redact the ID numbers after an election. This measure aims to prevent the identification of individual voters through their ballot codes, addressing the flaw identified by Springer's research.

How Ballots are Recorded

When a voter casts a ballot, a digital record, known as a cast vote record (CVR), is created. Each CVR is assigned a unique identifier by the tabulator. This identifier, if publicly accessible, could be correlated with other public records to potentially reveal a voter's selection.