Security researchers say a Georgia election-data design can permit a voter’s identity to be matched with the person’s ballot, a task made easier by modern AI tools. The dispute centers on identifying information, ballot images and the timing needed to certify and deploy a software update before the midterms. Georgia law promises ballot secrecy, while state officials say the proposed technical change lacks certification, funding and enough time for statewide deployment.
The reported sequence is specific. Georgia’s Constitution requires elections to use a secret ballot. State law also requires voting equipment to prevent another person from knowing how a voter voted. Researchers described a process in which records could be combined to associate a person with a ballot image. The same voting-equipment platform is used in jurisdictions outside Georgia, although local configurations and records practices differ. The current system prints a human-readable summary and a QR code used in tabulation.
The remaining factual record adds operational detail. Georgia lawmakers delayed a previously planned removal of QR-code tabulation until 2028. An agency lawyer told the State Election Board that the proposed update had not been certified by the secretary of state. The agency also said deployment would take months and the legislature had not appropriated the necessary funds. Recent state audits addressed tabulation accuracy; accuracy audits do not by themselves answer the separate ballot-secrecy question.
The surrounding context helps define what the report does and does not establish. Ballot secrecy and vote-count accuracy are different security properties. AI systems can accelerate optical character recognition, record linkage and pattern matching across large collections. A vulnerability does not prove that a malicious actor has exploited it in a live election. Changing certified election software close to voting can introduce operational and testing risks of its own. Primary state records establish official positions and audit results but do not independently resolve researchers’ technical findings.
Some points remain unresolved. No evidence was presented that voters’ choices had been systematically deanonymized in a live election. The feasibility and accuracy of linkage depend on access to specific records and their formatting. The next verifiable developments are State Election Board action on the proposed update. A documented certification and testing schedule for privacy changes.