Key facts
- AI can potentially link voters to their specific ballots in Georgia, compromising ballot secrecy.
- A Princeton University researcher used AI and public election records to identify voters' ballot choices.
- The researcher could recover the voting order for 1.52 million ballots in 114 of 139 examined counties.
- Georgia's outgoing secretary of state ordered redaction of ID numbers in tabulation data to address the vulnerability.
- Election officials are concerned about potential changes to voting processes so close to early voting.
- A previous flaw in Georgia's election software that connected voter identity to ballots was discovered about four years ago.
Researchers have warned that artificial intelligence could be used to compromise the secrecy of ballots cast in Georgia, potentially exposing voters' choices. Max Springer, a postdoctoral research fellow at Princeton University’s Center for Information Technology Policy, demonstrated how AI, combined with public election records, could identify individual voters' ballots. Springer stated that within hours, he had a system to analyze and identify secret ballots, and the AI agent indicated it could identify real voters' ballots with further information, without raising concerns.
This revelation has alarmed Georgia's election officials, particularly with early voting for the midterm elections, which will decide control of the US Congress, starting soon. Ballot secrecy is considered a crucial safeguard for voters against intimidation or undue influence. Springer's test indicated that the AI could recover the voting order for 1.52 million ballots, or 98.9% of in-person ballots in 114 of the 139 counties he examined. In smaller counties like Heard, the AI could match most early in-person voters to specific ballots.
This issue echoes a flaw discovered about four years ago in Georgia's election software that could link a voter's identity to their ballot. The current vulnerability stems from how voting machines assign unique identifiers to cast vote records (CVRs), which are essentially rows in a spreadsheet representing scanned ballots. While these identifiers were intended to prevent election workers from double-stuffing ballots, their predictability, when combined with other data like sign-in logs or camera footage, allows AI to link them to specific voters.
Georgia's secretary of state's office has reportedly sought funding for three years to upgrade the voting system and address vulnerabilities, but legislative appropriations have not materialized. Spokesperson Robert Sinners noted that requests for a complete overhaul of the software were met with "a big fat goose egg" from the legislature. Other states using similar Dominion software have either patched it or withheld vulnerable data from public view.
Georgia law presents a conflict between the constitutional requirement for a secret ballot and laws demanding transparency for election records, especially after revisions following the 2020 election disputes. An election board member proposed a solution involving poll workers holding ballots before they are scanned and shuffling them, but this was met with concerns from other officials about retraining election workers and potential physical mishandling of ballots. With early voting imminent, implementing significant process changes is considered impractical and insufficient.