AI Helps Reconstruct Georgia Voters’ Secret Ballots From Public Records

Researchers say artificial intelligence can help exploit a known weakness in Georgia’s voting system to potentially connect supposedly secret ballots with individual voters using publicly available records. The vulnerability concerns ballot secrecy, not the accuracy of the vote count.

The weakness involves electronic ballot records that can be reconstructed into the sequence in which votes were cast.

Those records can then be compared with other public information, including early-voting records, to narrow down who may have cast particular ballots.

Princeton researcher Max Springer tested how much AI changes the difficulty of doing this.

He gave a publicly available AI coding assistant the published research explaining the vulnerability.

The system helped perform the technical work needed to reconstruct voting patterns and, in many cases, narrow ballots to individual voters.

Springer says he required no access to voting machines, private databases or proprietary source code.

AI CHANGES WHO CAN EXPLOIT A VULNERABILITY

Georgia officials are taking interim measures, including scrambling the order of released ballot data and restricting some sensitive records.

A proposed requirement to install the voting-system vendor’s software fix before November was rejected amid concerns over certification, timing, funding and authority.

The vote count itself is not the issue.

The secret ballot is.

Ballot secrecy protects voters from intimidation, retaliation and vote-buying by preventing others from proving how an individual voted.

The Georgia vulnerability also exposes a much larger security problem.

Some technical weaknesses remain relatively contained because exploiting them requires specialist knowledge, significant time or both.

AI can dramatically lower that barrier.

How many systems are considered secure today simply because exploiting their weaknesses is too difficult for most people?

Sources

Associated Press: Georgia secret-ballot vulnerability

Princeton CITP: An algorithmic failure beneath the secret ballot