How Secure Are US Elections? Voting Machine Risks and the Growing Threat of AI

What it is

A SecurityWeek opinion-and-interview piece that examines in-person US voting, using ballot-marking devices, through a cybersecurity lens. It rests on quotes from vendors and security executives, not on new technical research. The article gives no CVEs, test results or commands.

How it works

Shai Bargil (Sequent) argues that Election Assistance Commission (EAC) certification is not tied to the four-year election cycle. Certification can start in any year and takes up to about two years. A machine certified early in a cycle could therefore run unpatched for two years or more. Bargil says every machine delivered to DEF CON's Voting Village has been hacked with hands-on access.

The counting scanners read a barcode or QR code, not the human-readable text. A voter cannot check what is actually counted.

What to do

The article offers design recommendations, not procedures:

• Christiaan Beek (Rapid7): treat voting machines as one component in a potentially hostile system. Assume vulnerable software and firmware, supply chain targeting, and a compromised admin account. Make compromise hard, limit blast radius, expose manipulation, and keep a way to reconstruct and verify the outcome.
• Bargil: make machine software open source, so outside researchers can review and pentest it alongside EAC certification. Have the machine send an encrypted vote record directly to the counting system, in addition to the paper ballot, so the count can be verified independently.
• Dave Gerry (Bugcrowd): put the harder defensive work on the perimeter, meaning voter registration databases, county websites, and email accounts of campaign staff and election workers. He says adversaries have focused there since 2016.
• Casey Erikson (NRLabs): constrain risk through access control, verifiable backups and layered redundancy.

Caveats

• Bargil's company sells verifiable voting software. His claims about certification and Voting Village results are his own, not independently sourced in the article.
• Microsoft ElectionGuard also uses homomorphic encryption of votes and is open source, but the article says it does not propose open-sourcing the voting machine software.
• Adam Marrè (Arctic Wolf) says most election-related activity targets public perception, not vote changes. The article's examples include the January 2024 New Hampshire Biden deepfake robocall, which failed.
• The article calls its machine-compromise scenarios implausible but possible. It states there is no evidence machine tampering has changed a US election outcome, and it cites pre-election testing, chain of custody, bipartisan poll workers and post-election paper audits as safeguards.
• Voter registration, voter ID and fraud are explicitly out of scope.