The source code you are not allowed to read
Every voting machine running your elections contains software you are legally forbidden to read — and the people certifying it as safe have often never read it either.
It is the autumn of 2007, and teams of computer scientists from the University of California are sitting in front of voting-machine source code they have never been allowed to see before. California's Secretary of State has just handed them something remarkable: the actual software running inside the Diebold, Sequoia, and Hart InterCivic systems certified for use across the state. The researchers have weeks. They have the code. They have the hardware.
What they find is not reassuring.
The Diebold report concludes that "the technological controls in the software did not provide sufficient security to guarantee a trustworthy election." The Sequoia report finds "significant security weaknesses throughout the system, raising serious questions as to whether the software could be relied upon to protect the integrity of elections." The Hart InterCivic report notes the system "was not designed on defense-in-depth lines" and that its components "largely trust one another."
These are not fringe findings. These are the conclusions of independent, credentialed scientists given the same access that certification labs had already used to call these systems fit for public use.
The difference? The certification labs said: approved. The independent scientists said: not so fast.
The story of election-technology secrecy is the story of that gap.
Why you are not allowed to read the software
Here is the standard arrangement in most U.S. jurisdictions — and many democracies worldwide.
A private company writes software. That software counts votes in public elections. The company considers the code a trade secret. It licenses the system to a government body, which runs your election on it. An accredited testing lab examines the code under a nondisclosure agreement. The lab issues a certificate. The election happens.
You voted. The machine counted. You cannot read the software that did the counting. Neither can your neighbor. Neither, in most cases, can the county election director who bought the system. Neither can a journalist, an academic researcher, or a judge — not without the vendor's permission, which is almost never granted.
You are asked to trust a process whose mechanism is a trade secret.
This is not a hypothetical complaint. It has concrete consequences — documented in court, in audit reports, and in peer-reviewed computer science.
What happens when researchers do get access
California's 2007 top-to-bottom review is the clearest case study in what closed code has been hiding. The Secretary of State gave independent teams source-code access to study the Diebold, Sequoia, and Hart InterCivic systems. The findings, now public, describe vulnerabilities to malicious software, viruses that could spread between machines, audit mechanisms that might not detect determined attacks, and network interfaces that were unsecured.
One year earlier, Princeton researchers had already demonstrated the worst-case scenario with a real Diebold AccuVote-TS machine. Ariel Feldman, J. Alex Halderman, and Edward Felten showed that an attacker with about a minute of physical access could install malicious code capable of stealing votes while keeping every internal log, counter, and record consistent. They built a working voting-machine virus that spread automatically between machines during normal election activity. The machine had been certified. The code had been reviewed. Neither fact stopped the researchers from doing what they demonstrated.
The Princeton team's paper has a simple, brutal implication: if the software is wrong — through malice or accident — and no independent party can read it, no one will know until something goes visibly wrong. And it may never go visibly wrong.
That is the core of the secret-source-code problem. Not a conspiracy. Not a claim that any particular election was stolen. A structural property of a closed system: undetectable errors are possible, and the mechanism for detecting them has been contracted away.
Ohio confirmed it in 2007. Its EVEREST review — nine weeks of work by teams from Pennsylvania State University, the University of Pennsylvania, and WebWise Security — examined systems from ES&S, Hart InterCivic, and Premier (formerly Diebold) and reached a verdict that applied to all three: each system possesses "critical security failures that render their technical controls insufficient to guarantee a trustworthy election." Shared failures included insufficient defense against insider attacks, improper use of cryptography, and audit logs that could be altered or deleted by the same people they were supposed to monitor.
Logs the administrators could erase. Audits that would not catch the thing they were designed to catch.
The certification trap
At this point a reasonable person objects: but these systems are certified. Doesn't that mean someone checked?
Yes. Someone checked. Once. Under conditions the vendor helped design. Using standards that, as federal auditors noted in a 2005 GAO report, were still being developed years after HAVA mandated them. The GAO found that "key activities remained incomplete" — that the machinery for assuring voting-system security and reliability was still under construction even as certified systems were being deployed.
Certification in election technology means: this software passed a defined set of tests, administered by an accredited lab, at the time it was submitted. It does not mean the same software is running today. It does not mean the tests covered every attack scenario. It does not mean an independent researcher cannot find a critical flaw the certification process missed.
It is a snapshot, not a standing guarantee.
"Certified" is what a vendor tells you. "Verifiable" is what you can check yourself.
The distinction matters most in close elections. When a race turns on a few hundred votes, and the system counting those votes cannot be independently inspected, you are not in a position to confirm the result. You are in a position to believe an assurance.
The Coffee County, Georgia episode of 2021 illustrates what assurance alone is worth. On January 7, 2021, a computer-forensics firm walked into the Coffee County elections office and copied the entire voting system — the Election Management System server, ballot-marking devices, the scanner, Georgia's statewide Dominion software — with the help of local officials. The software that was supposed to be secret, locked down, certified, and secure was reproduced and distributed in an afternoon. The Georgia Secretary of State later described it as unauthorized access in violation of state law. Several people connected to it pleaded guilty.
Secrecy, it turns out, is not a security strategy. It is a wish.
What the alternative actually looks like
On August 21, 2018, the California Secretary of State certified something that had never existed before: Voting Solutions for All People, the VSAP Tally Version 1.0, built by Los Angeles County. The state called it "California's first certified election system to use open-source technology" and "the first publicly-owned, open-source election tally system certified under the California voting systems standards."
Read that slowly. Los Angeles County — the largest voting jurisdiction in the United States — built and owned its own voting system. The code was public. The county established an open-source working group so other jurisdictions could build on it rather than buying another vendor's black box.
This is what the alternative looks like: not a foreign concept, not a theoretical design, but a working, certified system in the most populous county in America.
Open-source code does not mean every voter reads the software. It means that any qualified researcher — from any university, any country, any party's security team — can read it. It means that if there is a flaw, the whole world can find it, not just the vendor's internal team. It means that the fix can be verified too.
The contrast with closed systems is not abstract. When Swiss Post published the source code of its internet voting system in 2019, independent cryptographers Sarah Jamie Lewis, Olivier Pereira, and Vanessa Teague found something that internal testing had not caught: a trapdoor in the cryptographic shuffle proof that would let an authority with access to certain values generate a proof that appeared to verify correctly while covering altered votes. The flaw was invisible in a black box. It became findable the moment the source was public.
Swiss authorities suspended the system pending remediation. The publication of the code did not create the flaw. It exposed one that was already there.
A defect that can be found and fixed is safer than a defect that is hidden and trusted.
What courts have said
The secrecy problem is not only a computer-science argument. Courts have noticed.
Germany's Federal Constitutional Court, in its landmark 2009 ruling on electronic voting, held that the use of voting computers requires that "the essential steps of the voting and of the determination of the result can be examined by the citizen reliably and without any specialist knowledge of the subject." The Court found that because the Nedap machines stored votes only in electronic memory — with no independently verifiable record — they failed this standard. The public-verifiability principle flows directly from the constitutional requirement of public elections.
That ruling did not wait for proof of fraud. It said: a system whose correctness rests on hidden software is constitutionally insufficient, regardless of whether anyone has yet demonstrated an attack.
In the United States, a federal court in Georgia came at the same problem from a different angle. In Curling v. Raffensperger, decided October 2020, the court heard expert testimony about Georgia's Dominion ballot-marking devices and found that the system "does not provide a verifiable and auditable ballot record because it relies on the QR code for vote tabulation and that code itself cannot be read and verified by the voter." The court quoted the National Academies' conclusion that "no technical mechanism currently exists to ensure a vote-counting application produces accurate results and that testing alone cannot ensure systems have not been compromised."
Testing alone cannot ensure it. The court said so. This is not a partisan claim. It is a statement about the limits of certification.
The court, citing proximity to the November 2020 election, declined to order a switch to hand-marked paper ballots. That is a reasonable judicial decision about timing. It does not resolve the underlying problem.
The harder question: open source is necessary, but is it enough?
Here is where intellectual honesty requires a step past the easy argument.
VSAP is a genuine advance. Published code is a genuine advance. But open-source software and a working, verifiable election are not the same thing. The Switzerland case proves it: the source was published, but the cryptographic proof was written in a way that made the trapdoor hard to find. It took specialized mathematical expertise, not just a code reviewer reading functions.
Open code that is inspectable in principle but requires a PhD in cryptography to actually audit is not the same as open code that any competent developer can read.
Then there is the implementation gap. What is published and what runs on the machine on election day may not be identical. This is the reproducible-build problem: to verify that the software certified is the software deployed, you need to compile the published source and confirm the binary matches what is on the hardware. That is a step most jurisdictions do not take, and most vendors have not been required to support.
And even a correct, public, independently audited software stack still needs a durable paper record and a real audit to check it against. Georgia's 2020 statewide hand count — roughly five million ballots examined across 159 counties — affirmed the machine tally to within about a tenth of a percent. That confirmation was only possible because voter-marked paper existed. Without it, open source alone would still leave a gap.
The full answer is: open and inspectable source code, plus reproducible builds, plus voter-verified paper, plus mandatory independent audits, plus machine-readable precinct-level results published the moment they are final. Each layer catches what the others miss. Remove any one of them and you are asking someone to trust.
What is still not verifiable — and what would make it checkable
Right now, in most jurisdictions, here is what you cannot independently confirm:
That the software certified is the software running on the machines in your precinct. That the software running on the machines has not been modified since certification. That the compilation process produces the same binary from the source the vendor or county published. That the audit logs on those machines have not been altered. That a certified-but-flawed system has had its flaw found, because no one outside the vendor's team has been permitted to look.
None of these gaps require malicious actors. Each is a structural property of closed, proprietary systems.
What would make them checkable? Published, version-controlled source code updated each time the software changes. Reproducible builds, so any competent technologist can confirm the deployed binary matches the published source. Mandatory access for independent security researchers — not a closed lab under NDA, but open adversarial review, the model Brazil has used for its national electronic voting machines since 2009, explicitly inviting qualified testers to try to break the system before each election. Voter-verified paper records that are physically distinct from the machine's electronic output. And precinct-level results published instantly, in machine-readable formats, the moment each precinct is canvassed — so the public can aggregate and cross-check the numbers independently rather than waiting for an official total to trust.
The gap between where most systems are and where that standard sits is documented, specific, and closeable.
The first step is acknowledging that "certified" and "verifiable" are not synonyms — and that only one of them gives you something you can check yourself.
See how common the secret-source-code gap is across jurisdictions worldwide
Explore the full election-integrity gap map
Two-minute version of the verifiability problem
Sources
- California Secretary of State, Top-to-Bottom Review (2007) — decisions and independent experts' reports
- Source Code Review of the Diebold Voting System (University of California, 2007)
- Source Code Review of the Sequoia Voting System (University of California, 2007)
- Source Code Review of the Hart InterCivic Voting System (University of California, 2007)
- EVEREST: Evaluation and Validation of Election-Related Equipment, Standards and Testing — Final Report (December 7, 2007)
- Feldman, Halderman & Felten — Security Analysis of the Diebold AccuVote-TS Voting Machine (USENIX/EVT)
- Lawfare — What the Heck Happened in Coffee County, Georgia?
- Atlanta Journal-Constitution — Coffee County was the site of a 2021 election breach
- California Secretary of State news release (Aug. 21, 2018): certifying LA County VSAP Tally
- LA County Registrar-Recorder/County Clerk, VSAP official program report
- Lewis, Pereira, Teague — Ceci n'est pas une preuve (trapdoor commitments in the Scytl-SwissPost Internet voting system), 2019
- Bundesverfassungsgericht, Judgment of 3 March 2009, 2 BvC 3/07 and 2 BvC 4/07 (English translation)
- Curling v. Raffensperger, No. 1:17-cv-2989-AT, Opinion and Order (N.D. Ga. Oct. 11, 2020)
- GAO-05-956, 'Elections: Federal Efforts to Improve Security and Reliability of Electronic Voting Systems Are Under Way, but Key Activities Need to Be Completed' (Sept. 21, 2005)
- Georgia Public Broadcasting — Risk-Limiting Audit Confirms Biden Won Georgia
- Tribunal Superior Eleitoral (Brazil) — Teste Público de Segurança dos Sistemas Eleitorais