Truace
Subscribe

Georgia holds emergency meeting on AI exposing voters’ secret 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? Last month, Max Springer, a postdoctoral research fellow at Princeton University’s Center for Information Technology Policy, tested the prospect of identifying voters based on their Georgia ballots with a $20 subscription to an AI large language model and data obtained from an Open Records Act request. “Within a cou…

By the Truvace desk2 min read1 sourceJournalismTRV-2026-1476· cite & verify
Georgia holds emergency meeting on AI exposing voters’ secret ballots

In brief

A Princeton postdoctoral researcher showed that a subscription AI model, an early-voting list and county cast-vote record files obtained through an open records request could be combined to work out the order in which ballots were cast. In the counties he examined, this allowed most in-person ballots to be matched to the sequence of voters. Georgia's state elections board met in an emergency session to discuss the exposure.

The finding matters because ballot secrecy protects voters from pressure and intimidation, and it surfaced shortly before early voting in a midterm election. Officials disagree on the remedy, and the article leaves unresolved whether any fix can be adopted in time without creating new problems or conflicting with transparency law.

Main points

  1. The researcher said he built an analysis pipeline within hours and that the AI agent never refused or raised concerns about the exploit.
  2. He recovered voting order for 1.52m ballots, or 98.9% of in-person ballots, in 114 of 139 counties examined.
  3. In Heard county, the agent matched most of 650 early in-person voters to a specific ballot and the rest to within a single swap.
  4. Secretary of state Brad Raffensperger ordered ID numbers redacted from public releases after an election, which a cryptographer said substantially addresses the flaw.
  5. A proposal to shuffle printed ballots before scanning was criticized as burdensome and as creating new risks.

The problem

A researcher using a cheap AI model and public records recovered the voting order for most in-person ballots in 114 of 139 Georgia counties, which could link voters to their secret ballots.

The rundown

The weakness lies in the unique identifier the tabulator assigns to each cast vote record, which is not random enough. If ballot order is paired with other data such as sign-in records or camera imagery showing when someone entered a building, a ballot can be linked to a named voter.

The identifier exists to deter fears of double-stuffing ballots through tabulators. Election security researchers had found a related flaw about four years earlier, and the secretary of state's office says it has sought money for an overhaul for three years without success amid a political feud.

Georgia law requires both a secret ballot and transparent election records. Board member Salleigh Grubbs argued the redaction fix violates the law, while other members and county officials objected to her tray-shuffling proposal because it would require retraining and could introduce new errors. The article says other states have patched the software or withhold the vulnerable data.

Sources

  1. JournalismThe Guardian2026-10-02

The debate