Encrypt the ballots, then audit in public: Ginny Badanes on ElectionGuard
Ginny Badanes directs strategic projects for Microsoft's Defending Democracy program, and for a couple of years her team has been building ElectionGuard, an open source project for end-to-end verifiable elections. The promise is simple to state and hard to build. A voter can follow their own ballot all the way through the count without anyone ever learning how they voted, because the math works on encrypted ballots.
Earlier this year ElectionGuard had its first real test, a pilot in Wisconsin run with VotingWorks, a nonprofit voting vendor. Ginny called it a proof of concept and hopes it leads to more pilots and, eventually, to the technology being built into many different voting systems. But what I wanted to hear about was the newer idea, which joins ElectionGuard to audits.
Her team follows what the top election security experts say, and one of the things they say most often is that the best protection is an audit after the fact. Any audit is better than none. A risk-limiting audit is a more efficient kind: you check a random sample of paper ballots against the reported results, and keep going until the sample gives you very high statistical confidence that the reported outcome is right. Colorado has run them statewide since 2017.
Risk-limiting audits have their own tension, though. Some jurisdictions publish the electronic record of every ballot cast, so anyone can see exactly which ballots were pulled for the audit. Others worry about what that exposes. A ballot marked in an unusual pattern can identify the voter who marked it, which opens the door to coercion.
What if you encrypted those ballots, put the encryption out publicly, identified which ones you wanted for the audit from the encryption, and then at that point reveal the results of the ballot for the purpose of the audit?
Engineers will recognize the design. Publish a commitment first, so nobody can change it later. Choose the sample in public, from the committed record. Reveal only what the audit needs. Everyone can check the process, and nobody learns more than the audit requires.
I said that risk-limiting audits looked to me like the one approach being widely adopted. Ginny corrected me gently: widely agreed upon, not yet widely adopted. She did not want to rank them above other audits either. Some election offices are not set up for that particular method, and the experts she listens to tend to say that auditing is good, risk-limiting audits can be better, and it depends on the setup.
The point she kept returning to was timing. An audit happens after the votes are cast and before the result is certified, and that is the window this year's public is about to discover. Election results are not done on election night. Even in a landslide it is not officially over; the chance the final count differs from election-night reporting may be small, but the election is not done until it is certified. That gap is where she wants a real validation step, not just counting and certifying.
This year may be election week rather than election day in many places. Counting that many paper ballots is messy and stressful, with a lot of technology, a lot of people and a lot of decisions in a short time. Paper you can recount, an audit you can watch, and a record you can check without exposing anyone's vote are what turn that mess into a result people can trust.
From Lincoln Shorts, the interview series I hosted for Lincoln Network, recorded 1 September 2020.
