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? Continue reading...
A Princeton researcher discovered that publicly available election records combined with AI could potentially link voters to their ballots. During an AI test, Max Springer used a $20 subscription to an AI large language model and data obtained from an Open Records Act request. Within a few hours, Springer created a pipeline to analyze and identify secret ballots across the state of Georgia, revealing the agent's ability to identify real voters' ballots without hesitation.
Election secrecy is crucial for protecting voters from potential influence or intimidation, allowing them to cast their vote freely. However, Springer's test raised concerns among Georgia's election officials, particularly less than two weeks before early voting commenced in the crucial midterm election that would decide control of the US Congress.
Georgia's state elections board convened an emergency meeting to address the issue of AI's potential to identify voters through their ballot code. Ben Adida, an MIT-trained cryptographer and founder of the election technology non-profit VotingWorks, noted that Georgia's outgoing secretary of state, Brad Raffensperger, secured tabulation data by ordering any public release to redact ID numbers after each election, which he believed addressed the vulnerability.
However, election workers create a record of every voter who enters a polling place, and their choices are printed on a paper ballot, which is then scanned and tabulated.
The flaw lies in the machine's recording of the ballots, which is not sufficiently random. By utilizing AI, Springer could recover the voting order for 1.52 million ballots, or 98.9% of in-person ballots in 114 of the 139 Georgia counties he examined. In smaller counties, such as Heard with few voters, the agent could match the majority of 650 early in-person voters to a specific ballot, or a single swap.
The issue of ballot identification has been known for years, with election security researchers alerting Georgia's election officials about a flaw in their software four years ago. Despite multiple requests for funding to address the problem, Raffensperger and conservative legislators have largely failed to prioritize the issue, resulting in no funding being allocated.
Written by urgent.news from Guardian Technology's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.