If We Voted on a Blockchain, Nobody Could Tamper With It — Right? Make It Checkable, but Provable to No One

The Secret Ballot Was Introduced to Make Verification Impossible

In the 1880s and 1890s, nearly every state in the U.S. adopted a single voting method: identical government-printed ballots, marked behind a curtain where no one could see (DOI: 10.36646/mjlr.36.3.absentee). The goal of this reform, called the "Australian ballot," was to cut off vote-buying and coercion by party organizations.

A secret ballot is a method that ensures no one other than the voter can know who voted for whom. A person who buys a vote cannot verify whether the vote they bought was actually cast, and a coercer cannot know if their threats worked. The secret ballot is a technology introduced to make verification impossible. "Unverifiability" was one of democracy's oldest security devices.

Now, however, people are demanding the exact opposite: they want to be able to verify whether their vote was recorded as cast, and counted as recorded. The answer most frequently attached to that demand is, "If you put it on a blockchain, doesn't that mean nobody can tamper with it?"

A blockchain is a ledger linked in a chain where every batch of records includes the seal value of the preceding batch—a short value that changes even if a single character is altered—making it so that modifying a single line after the fact creates a mismatch that stands out ("What Is a Blockchain"). Representative responses from computer security research strongly object to the idea of protecting elections with this device. The reasons are not singular, and the most important reason has almost nothing to do with blockchains.

Two things must be clarified in advance. First, "vulnerabilities exist" and "fraud occurred" are completely different statements. Everything in this article consists of technical vulnerabilities in systems reported by researchers in academic literature. Several of the cases below are instances where researchers responsibly notified the respective institutions and companies and jointly distributed mitigations.

Second, when this distinction breaks down, there is a measurable harm. There is a pre-registered survey experiment from the 2024 election period that asked the same respondents the same question twice with a time lag. Pre-registration is the practice of publicly disclosing an analysis plan before running an experiment. In that experiment, news coverage of cyberattacks significantly reduced the belief that elections are conducted with integrity, and this effect crossed party lines. Even news about cyberattacks unrelated to elections undermined trust in voting systems (DOI: 10.1007/s11109-026-10147-6). That is why this article talks only about design.

The conventional wisdom has a basis. There have been continuous attempts to map the properties championed by blockchains—namely, that records cannot be altered secretly, that anyone can view them, and that there is no central administrator—one by one onto the security requirements of electronic voting. One representative response in this field has a title that is essentially its conclusion: "From Bad to Worse: Internet Voting and Blockchain in Elections," published in the Journal of Cybersecurity (DOI: 10.1093/cybsec/tyaa025).

This paper, which compiles and reviews previous research, judges that claims of internet and blockchain voting improving election security are unsubordinated and misleading. At the level of computer security at the time the paper was written—specifically, as long as standard attacks like malware work—even if online voting increases turnout, that increase comes at a cost: the loss of meaningful guarantees that votes were counted as cast and were neither changed nor discarded undetected. The authors also note that research on actual turnout effects remains inconclusive. The authors view that this risk not only persists in blockchain-based systems, but that blockchains can introduce additional problems. The authors' final judgment is that internet and blockchain voting significantly increase the risk of undetected, nation-scale election failure (DOI: 10.1093/cybsec/tyaa025).

Why they go that far comes down to two twists, both of which have almost nothing to do with the properties of the ledger.

If You Can Prove It, You Can Sell It

Electronic voting simultaneously requires two properties that push against each other. One is the secret ballot we looked at earlier. The other is verifiability—the property of being able to confirm that my vote was recorded as cast and counted as recorded. A voting method that reveals nothing at all keeps secrets but cannot verify the results, and opening everything up leaves no secrets.

Let us picture someone just walking out of a polling place. Standing outside the door is the person who handed him money yesterday. Under the Australian ballot, all he can offer is his word, and the person who paid him has no way to check whether that word is true. Now let us suppose he is holding a confirmation slip that reads, "Your vote has been recorded for candidate No. ○." To him, that piece of paper is reassurance that his vote went in properly; to the person outside the door, it is proof that he got his money's worth. The exact same piece of paper does both jobs.

At this point, "Zero-Knowledge Proofs: Proving the Truth Without Revealing the Data" might come to mind. A zero-knowledge proof is a device that convinces someone else that you know a secret while revealing nothing about the secret itself (「영지식 증명 — 정보를 공개하지 않고 참임을 증명하는 법」). However, a zero-knowledge proof lets you prove something to someone else. What we need in an election is one step stronger. Voters must be able to confirm for themselves that their vote was properly reflected, but must not be able to prove it to anyone else. The moment you can prove it, you can also prove it to vote buyers and coercers.

There are two names for this requirement. Receipt-freeness is the property that prevents voters from proving to others how they voted, even if they want to. Coercion resistance is the property that prevents someone from forcing a specific vote through threats or rewards, even if they try to compel it (arXiv:2604.00188, preprint).

This tension is also called the "verification paradox." Voters must be able to check whether their vote was included, but the moment that confirmation information becomes evidence they can hand to a third party, it provides a basis for post-election intimidation. Researchers who coined this term in 2026 built a prototype on a blockchain restricting participants to pre-verified locations, and conclude that it shows the potential to partially alleviate the paradox. However, the authors themselves disclose limitations: there is no formal security proof and there is only a single evaluation environment (DOI: 10.3390/app16146953).

There is a theoretical model calculating how secret ballots change vote-buying. Negative vote-buying is buying abstention—that is, paying people not to vote. This model calculates that while negative vote-buying is possible under open voting, it is never optimal, whereas under a secret ballot, mixing positive and negative vote-buying becomes optimal. The approach is to pay lukewarm supporters to show up at the polls, and pay lukewarm opponents to stay home. The conclusion is paradoxical: introducing a secret ballot can increase the proportion of voters acting against their own preferences and can even lower the cost of buying an election (DOI: 10.2139/ssrn.1706490, preprint). This model is a preprint uploaded to SSRN, a paper-sharing site, meaning it has been published without verified peer review.

Secrecy is not just a matter of technology. In a survey experiment conducted in Mexico, just receiving material goods from a candidate reduced voters' belief in the integrity of their vote. The authors model vote-buying as a signaling game. Voters do not know in advance whether politicians have the ability to monitor their behavior, and the politician's act of handing over money or goods is interpreted as information about that capability (DOI: 10.1017/psrm.2024.54). The authors suggest that under certain conditions, simply handing over goods shatters the certainty of the secret ballot, making vote-buying effective. Even if the system keeps secrets, coercion can still work if voters do not believe they are being kept.

Thus, if verifiability is poorly designed, it reopens the door to the market for votes that the 1880s reforms sought to block. Furthermore, most of the verification discussed in this article takes place outside the polling place. As a legal paper addressing the tension between absentee and secret voting puts it, a vote cast without drawing the curtain is potentially exposed to coercion, vote-buying, and fraud (DOI: 10.36646/mjlr.36.3.absentee).

There have even been cases where devices meant to protect secrecy broke it. Finland introduced postal voting as a new voting method for overseas citizens and included a witness requirement to protect voting secrecy, and one study looks at how that was implemented in the 2019 election. The analysis of the experiences of 31 Finnish citizens abroad produced unexpected results. Many voters actually kept their secrets well, but failed to convincingly demonstrate that fact to their witnesses. For them, the witness requirement only functioned to break a "secondary secret"—the secrecy of the very fact that they had voted (DOI: 10.3389/fpos.2021.630001). A mechanism put in place to protect secrecy ends up breaking secrecy on another level.

Three Remaining Difficulties Even When Protecting the Ledger

The second twist is simpler. The integrity of the ledger that blockchains protect was never the bottleneck to begin with.

Three things are difficult in an election: verifying from a distance who is eligible to vote, checking whether the device held by the voter is honest, and making it possible to verify the results without trusting the software. What "From Bad to Worse" confirmed by gathering previous research was that the security risks of online voting remain intact in blockchain-based systems (DOI: 10.1093/cybsec/tyaa025).

The third has a precise name. It is software independence, defined in a paper written to help create U.S. voting system certification guidelines. "A voting system is software independent if an (undetected) change or error in its software cannot cause an undetected change or error in election results." This means that even if the software secretly goes wrong, something outside the software remains available for verification so that the error cannot hide inside the results. Optical scan methods and some cryptography-based systems fall under this category. This paper suggests favoring software-independent systems and avoiding software-dependent ones (DOI: 10.1098/rsta.2008.0149).

Blockchains do not grant this property. A digital signature is a mark that can be created only by someone holding a private key—a key that only the owner keeps and never sends anywhere—and that anyone can verify using its paired public key ("Keys and Locks"). If the voting app on my smartphone generates a vote different from my intention and signs it with my key to submit it, that vote is honestly recorded on the ledger. The signature only says that the vote came from my key. As "What Is a Blockchain" noted in the section "What a Seal Does Not Guarantee," a seal does not guarantee that the written contents are true. The ledger protects "who submitted what," but does not protect "was what was submitted that person's intention."

A doctoral dissertation that systematically surveyed the market and reviewed academic literature together records that most industrial internet voting systems are effectively black boxes and provide zero meaningful verifiability. The same survey also found a few systems that actually implemented cutting-edge techniques, and the author writes that this shows those techniques are usable in practice. In response, the author proposes a code voting method where voters input a very short code instead of a candidate's name. This is intended to allow secret voting even in situations where the voter's own device has already been compromised (DOI: 10.70675/a5792297ze7fez462dz87c0ze0025ac20a95).

The first difficulty, remote voter authentication, also remains. A paper examining the introduction of internet voting under martial law in war-torn Ukraine explicitly lists the problem of remote voter identification on its list of major threats. It is a threat placed alongside infrastructure cyber vulnerabilities, external interference, violations of the secrecy of the vote, and the undermining of the legitimacy of results (DOI: 10.31861/mediaforum.2026.18.98-119). None of these are problems solved by the ledger's data structure.

Secrets Leak Next to Passwords

Even if a password holds firm, secrets can leak outside the protocol—the procedures systems use to communicate with one another. A side channel is a passage through which information leaks, not from the contents systems natively exchange, but from the traces left beside them, such as data length or the frequency and timing of records. The first two cases below leaked through this passage, while the latter two are about receipt code design and paper slips.

The first passage is the confirmation screen. Online voting systems typically display a screen confirming a choice right before submission. One study asked whether someone watching the network could figure out a voter's choice based solely on the "length" of the exchanged data. Analyzing a commercial implementation in detail, they found that even though the data length varied randomly due to dynamic page content and compression, in some cases they could guess the voter's choice with 100% accuracy. The authors coordinated with the vendor to disclose the vulnerability and jointly release mitigations, and added that this discovery was possible only because the vendor provided a public demo—and such vendors are rare in the industry (DOI: 10.1007/978-3-031-15911-4_3).

The second passage is national electronic identity logs. Estonia's internet voting includes revoting as a legal safeguard against coercion. Revoting allows an electronic vote to be cast multiple times during the advance voting period with only the last one counted, and the number of times it was cast must remain secret. One can cast a vote as instructed in front of a coercer, and later cast it again alone.

In Estonia, voters verify their identity using a state-issued electronic identity (eID) and digitally sign their ballot. However, under the eID auditing practices at the time of that paper, all signature records—including individual votes—were kept irreversibly and displayed to the user. This log revealed the exact number and timing of interactions the voter had with the voting system. The secrecy of the revote was broken, and the coercer could figure out whether the victim changed their choice. The paper compared alternatives such as continuously filtering logs, a separate identity token used exclusively for voting, and offline signatures, and responsibly notified Estonian authorities of the discovered vulnerability. The lesson drawn by the authors is this: well-crafted parts made with good intentions can meet and break coercion resistance through metadata side channels, and when an election system relies on external digital infrastructure, risk assessments considering the combination are essential (DOI: 10.1007/978-3-032-05036-6_12).

Not a single cryptographic element was broken. What broke was the boundary between the two systems.

The third case is the receipt code. An internet voting protocol used by Norway in the 2011 local elections and 2013 general election sent confirmation codes separately via mail and text message. A paper re-examining this points out three things. If the ballot box and the receipt generator collude, the secrecy of the vote is broken. A voter who keeps both codes can reveal their ballot even after the election is over. If codes are generated incorrectly, votes can be tallied for a different candidate without being detected. The authors attached solutions to each item, and summarized that while revoting is a fairly good solution for coercion or cover-ups, voluntary disclosure by choice still remains a problem (DOI: 10.1002/sec.1678).

The reason voluntary disclosure remains a problem lies in the definition of receipt-freeness. Receipt-freeness means that even if a voter wants to, they must not be able to prove how they voted. If voters can prove it when they want to, the grounds to refuse demands to "prove it" disappear. If there is a confirmation slip in the hand of that person in front of the polling station, the person who paid them can demand to see it.

Paper slips are not magic either. Tamper-evident seals are attached to voting machines and ballot boxes. A study reviewing the use of seals in New Jersey, USA, finds that these seals can usually be easily defeated due to how they are typically made and used, and that the effectiveness of a seal depends on the procedure for attaching and inspecting it. It concluded that New Jersey's seal operational procedures were not effective at all (DOI: 10.1145/2019599.2019603). This is a report on procedural and equipment vulnerabilities, not a claim that any election's outcome was actually changed.

What Paper, Statistics, and Procedures Have Achieved

There are also things this field has actually accomplished.

The order matters. First, you leave a physical record that does not require trusting software, which is a paper ballot. Next, you check that record statistically. The name of the second step is a risk-limiting audit (RLA). To translate the definition precisely: if the reported election outcome turns out to be wrong when every paper ballot is counted by hand, an RLA has a known minimum probability of escalating to a full hand count (arXiv:1809.04235, preprint). It is an audit where you draw a sample, check it by hand, and if you see signs that the result might be wrong, you move on to recounting everything. When the result is correct, you hand-check a much smaller number of ballots than a full recount and finish.

This audit has been run in real elections. Pilot audits were conducted in four elections across three California counties in 2008. The authors note limitations, stating that the approach at the time was too complex to become routine on a large scale, and cite a logistics bottleneck where commercial election management systems struggled to export data in a format suitable for audit calculations (arXiv:0905.4691, preprint). Since then, methods have continued to be refined, and PROVIDENCE, a round-by-round audit, was actually used for the tally audit by the Rhode Island Board of Elections in 2021 (arXiv:2210.08717, preprint).

The effect of audits does not end with statistics. In a pre-registered survey experiment presenting virtual scenarios to 2,000 Americans, how the audit was conducted mattered more than what the audit found. Structural factors, such as who carried it out and how the results were announced, had a larger impact on voters' evaluation of the outcome than the magnitude of any discrepancies actually discovered. Across party lines, respondents generally agreed that audits are an effective tool for catching counting errors (DOI: 10.1093/poq/nfae029). The authors suggest that election officials can maximize transparency in running audits to strengthen voter trust.

There are also achievements on the cryptographic side. End-to-end (E2E) verifiable voting is a voting method that allows voters to confirm their ballots were recorded and included in the tally while keeping the contents of their votes hidden. The November 2014 election in the Australian state of Victoria was the world's first deployment of an E2E verifiable electronic voting system at polling stations in a state-level statutory political election. The same election was also the first time visually impaired voters cast a completely secret ballot in a verifiable manner. The code is open source and the election outputs are verifiable. The number of votes received from the voter group targeted by this system was 1,121 (arXiv:1504.07098, preprint).

For visually impaired voters, it was the first time casting a completely secret ballot in a verifiable manner—meaning possessing verifiability and a secret ballot at the same time.

There are also criticisms of E2E verifiability. Systems that rely on complex cryptography become yet another black box for voters, and end up demanding blind trust in opaque software or outside experts, failing to build genuine public trust. Starting from this observation, a 2026 preprint puts forward software-free verification as a benchmark. It is the criterion that voters must be able to check election integrity on their own without relying on any software. The proposed method is a non-cryptographic approach that binds a machine-generated pseudonym hidden among other real ballots into a physical receipt and a hierarchical public ledger, making it auditable using basic arithmetic alone (arXiv:2603.21833, preprint).

The most sophisticated answer on the side of coercion resistance is fake credentials. A credential is a digital key that shows a person is qualified to vote. A fake credential looks identical to a real one so that it can be handed over to a coircer, but any vote cast with it is quietly dropped from the final tally (arXiv:2404.12075, preprint). The problem was how to issue it.

TRIP, a registration component of Votegral, solves that problem with a kiosk inside the physical voting booth. Voters print both real and fake credentials on paper inside the booth where no one else can see. You can check which one is real inside the booth, but afterwards the two are indistinguishable to other people's eyes.

Each paper credential contains an interactive zero-knowledge proof exchanged between the proving party and the verifying party, and that proof is sound only for the real one. Following the definition in "Zero-Knowledge Proofs: Proving the Truth Without Revealing the Data", soundness is the property that if a statement is false, a dishonest prover has a very small probability of tricking the verifier. Voters simply see the difference with their own eyes through the order of steps during the printing phase and do not need to understand the technical details. In fact, only coerced voters—or in the authors' phrasing, "the cases one hopes are rare"—need to trust the kiosk. In prototype experiments, the issuance delay was at most 19.7 seconds, and the authors report that the combination of coercion resistance and verifiability is supported by a formal proof (arXiv:2202.06692, preprint).

The Demand for Remote Voting Is Real

"It is dangerous, so let's not do it" is not an answer. The demand for remote voting has a face. The people highlighted in the previous Ukrainian paper include soldiers, internally displaced persons, citizens living abroad, and residents of affected regions (DOI: 10.31861/mediaforum.2026.18.98-119).

There are also cases that have been running for a long time. Estonia has operated internet voting since 2005, and an early report details its benefits. It states that while going to a polling station to vote takes about 44 minutes, electronic voting takes 6 minutes (DOI: 10.1007/s12290-013-0261-7). The literature is divided regarding voter turnout. That early report noted that turnout increased (DOI: 10.1007/s12290-013-0261-7), but a later study concludes that the increase in voter turnout was negligible in countries that introduced internet voting (DOI: 10.1002/poi3.160). Instead, something else was observed. In data from five nationwide elections in Estonia between 2009 and 2015, e-voting was very "sticky." People who voted via the internet once were very likely to continue doing so afterwards, and that rate was consistently higher than the rate at which paper voters continued paper voting (DOI: 10.1002/poi3.160).

A study that legally reviewed the experience of the same country is much more critical. Analyzing the 2023 Estonian general election as a case study, this paper points out the lack of independent observers to watch the internal operations of the electronic voting system and significant flaws in the appeal procedure. It reports that high-quality audits are needed at every core stage of end-to-end verification and result confirmation, and tentatively concludes that due to insufficient security guarantees, it is inappropriate to completely replace traditional paper voting with remote voting, at least in the near future (DOI: 10.21564/2414-990x.166.310197).

Some researchers flip the frame. Against the backdrop of Bulgaria's repeated early general elections and concerns over vote-buying, a preprint proposing reforms sums it up this way. The correct response is not to test, filter out, or exclude voters, but to redesign the procedure so that vote-buying transactions become unverifiable to the buyer while keeping the results verifiable to everyone. Its proposed tools include E2E verifiable voting featuring receipt-freeness and coercion-resistance, voter-verified paper records, randomized candidate ordering to neutralize instructions such as "vote for number X," and risk-limiting audits. This paper explicitly rejects imposing knowledge or "competence" tests as voting conditions and any measures that break the secret ballot. The reason is that they violate universal suffrage and are disproportionately harmful to the elderly, disabled, and minority communities (DOI: 10.2139/ssrn.7159338, preprint).

This frame aligns precisely with the axis of this article. What we must ask is, "Who should be able to verify, and who should not be able to verify?" Whether to allow verification is a secondary question.

Is Revoting the Answer? — The Coercer's Tools Evolve Too

Revoting is an elegant design that softens coercion, but the eID log study we saw earlier took aim at that very design. There is an even more fundamental result. The Loki protocol is a design that removes the burden on voters to store and hide pre-committed secret credentials, attempting to block last-minute coercion through revoting. One study found two vulnerabilities here. One is a brute-force attack, which neutralizes avoidance strategies and allows an attacker to cast a vote on behalf of the victim. The other is a forced-abstention attack, which lets an attacker figure out whether a victim followed instructions not to vote.

The authors generalize the first attack to draw out a dilemma. You cannot prevent last-minute coercion without a pre-committed secret credential (arXiv:2604.00188, preprint). The authors show that reverting to pre-committed secret credentials fixes both vulnerabilities, and they discuss the trade-off of tallying efficiency and stronger trust assumptions as the price. And the pre-committed secret brings back the storage and hiding burdens that Loki tried to lift. This round trip is the shape of this field.

The coercer's tools evolve too. A 2024 study points out that existing anti-coercion techniques underestimate the capabilities of malicious actors because they fail to consider how the technical tools available at the time can aid coercion and vote-selling. The research team actually demonstrates several ways for a coercer to enforce compliance or for a voter to prove their vote using blockchains, time-lock cryptography, privacy-preserving smart contracts, and trusted hardware. A smart contract is a ledger that records conditions and actions to be taken under those conditions, executing them automatically without human judgment ("What Is a Smart Contract?"). Some of those attacks succeeded against schemes that had been claimed or proven to possess receipt-freeness or coercion resistance. The conclusion is firm: our coercion models must be reevaluated, and the very definitions of coercion and receipt-freeness must be rewritten (DOI: 10.1007/978-3-031-72244-8_9).

As "Zero-Knowledge Proofs: Proving the Truth Without Revealing the Data" noted, when models and assumptions change, the conclusion changes too. Saying that receipt-freeness is proven is true only within the coercer model that the proof assumed, and as coercer tools grow, that model becomes obsolete. The fact that blockchain appears as a tool that aids coercion rather than prevents it in this study is especially ironic in light of the question of this article.

After definitions, the human side of the problem remains. Regarding whether ordinary people can actually understand and use fake credentials, there is the first systematic study to recruit 150 people for a mock election in Boston, USA. Among the 120 people exposed to fake credentials, 96% understood their purpose and 53% responded that they would create one if it were a real election. However, 10% accidentally voted with a fake credential. Among the same 120 people, 22% had directly experienced or closely known about coercion or vote-buying. They evaluated the coercion-resistant system as trustworthy to practically the same degree as hand-marked paper ballots at a polling place. The authors conclude that while supporting both the importance of the coercion problem and the potential of fake credentials, the user error rate remains a crucial challenge (arXiv:2404.12075, preprint).

This 10% is not light. In a coercion-resistant design, accidentally using a fake credential means that one's own vote is quietly left uncounted.

Before the Summary Card, Most People Did Not Verify

The most uncomfortable field measurement in this topic came from the human-verified link.

The chain of a verifiable system runs like this: voters check a paper record, that paper serves as the basis for an audit, and the audit guarantees the result. If the first link breaks, everything after it spins idly.

An experiment conducted at two primary election sites in Tennessee, USA, in 2018 tested that first link. The target was a voting machine that prints a summary card when a voter makes a selection on a touchscreen. The summary card is a paper printed with only the voter's choices extracted, and this goes into a scanner to become a vote. The question was simple: can voters review and confirm that the choices they made just minutes earlier were accurately recorded on that card?

The result was close to "no." In the preliminary evidence reported by the authors, most voters in actual polling place environments did not attempt to verify their summary cards even when instructed to do so. When exit-interviewing voters who did attempt a review, a statistically significant proportion failed to remember important details of the ballots they had cast just minutes earlier. They failed to notice errors on the presented paper, and they even failed to notice the fact that the presented paper was not the one they had cast. The implication drawn by the authors is heavy. Voters who do not securely leave their choices behind by directly marking paper by hand largely fail to detect tampering when a voting machine is compromised and votes are flipped (DOI: 10.2139/ssrn.3292208, preprint). This study is a preprint uploaded to SSRN, and the authors themselves call the results "preliminary evidence" and note that they generally align with other studies on memory.

There are also attempts to make verification easier. There is an attempt that actually layered a verification mechanism onto a live, commercial online voting system. Across two pilots, respondents found this form of verification easy to use and generally welcomed it. As a result, even the election organizers running the system could not alter the results without being detected (arXiv:1912.00288, preprint). However, what this pilot measured was the evaluation of ease of use, and it did not measure whether voters actually ended up verifying more.

What Makes Verification Possible

We summarize in three sentences.

First, verifiability and the secret ballot collide head-on. If you can prove your vote to someone else, you can also prove it to a vote buyer and a coercer. Therefore, what voting needs is a step stronger than "proving without disclosing"—it is "checking for myself while being unable to prove it to anyone." This is receipt-freeness and coercion resistance, and most of the difficulty in this field lies right here.

Second, the integrity of the ledger guarded by a blockchain alone does not solve this difficulty. The bottlenecks are voter authentication, the voter's device, and software independence, which makes it possible to check the results without trusting the software. Even if the ledger is split across multiple places, these three remain unchanged, and "From Bad to Worse," which gathers and organizes previous research, views that blockchains can rather add new problems. The partial mitigations of prototypes on blockchains also came from the devices added on top of the ledger rather than the ledger itself.

Third, what has actually been achieved lies on the procedural side, not the ledger side. It is the physical record that makes it unnecessary to trust the software, risk-limiting audits that inspect that record under a guaranteed level of probability, and who performs and how they announce those audits. In virtual scenario experiments, what moved voter evaluation more was how the audit was conducted rather than what the audit found.

The remaining problems are also clear. The demand for remote voting is real and has a face. Estonia has been running internet voting since 2005, and that case also revealed that coercion-mitigation mechanisms can be broken by logs well-made for other purposes. There is a dilemma that final-moment coercion cannot be prevented without secrets promised in advance, and the tools of coercers keep newly emerging. Above all, at least in front of the summary card printed by the ballot-marking device, most voters did not even attempt verification (preliminary evidence).

Reading this list as "therefore, voting cannot be trusted" runs counter to what these literatures say. The existence of vulnerabilities is not evidence of fraud. These papers are records of voluntarily finding and disclosing the weak points of their own field to figure out what needs to be fixed, and several were patched as soon as they were disclosed. How to prevent threat reports from having the effect of lowering trust has also been confirmed by experiments. Participants who watched a short video explaining the election's safeguards 4 weeks in advance maintained their trust even when they later encountered threatening reports (DOI: 10.1007/s11109-026-10147-6).

The question of this series was singular. Can something be verified without trusting an intermediary? Voting gives the hardest answer to that question. Making it verifiable, yet impossible to prove to anyone. Holding onto these two demands at the same time is the entirety of this field, and a single ledger yields neither of them. What is yielded is paper, statistics, procedures, and who watches over them.

Source Papers

• "Going from bad to worse: from Internet voting to blockchain voting" (2021) — DOI: 10.1093/cybsec/tyaa025 — Internet and blockchain voting significantly increase the risk of undetected national-scale election failures, and blockchain can introduce 'additional' problems.
• "On the notion of 'software independence' in voting systems" (2008) — DOI: 10.1098/rsta.2008.0149 — The definition of software independence and the proposal that such systems should be preferred.
• "Anonymous E-Voting on Hyperledger Fabric: Practical Mitigation of the Verification Paradox and Coercion Resistance" (2026) — DOI: 10.3390/app16146953 — The naming of the 'verification paradox'. The moment confirmation information becomes evidence for a third party, it becomes the basis for coercion.
• "The Absentee Ballot and the Secret Ballot: Challenges for Election Reform" (2025) — DOI: 10.36646/mjlr.36.3.absentee — The Australian ballot was introduced in the 1880s and 1890s to prevent vote buying and coercion, and votes cast without a curtain lack that protection.
• "Negative Vote Buying and the Secret Ballot" (2010) — DOI: 10.2139/ssrn.1706490 (preprint) — Under a secret ballot, combinations of positive and negative vote buying are optimal, which paradoxically can increase the volume of vote buying.
• "Does vote buying undermine confidence in ballot secrecy? Theory and experimental evidence" (2024) — DOI: 10.1017/psrm.2024.54 — Merely receiving goods reduces confidence in ballot secrecy (survey experiment + signaling model + laboratory experiment).
• "Practicing Ballot Secrecy: Postal Voting and the Witness Requirement at the 2019 Finnish Elections" (2021) — DOI: 10.3389/fpos.2021.630001 — The witness requirement for protecting secrecy instead operates in a way that breaks the secrecy of the fact that one 'has voted'.
• "Formal Methods for Internet Voting" — DOI: 10.70675/a5792297ze7fez462dz87c0ze0025ac20a95 — Most industrial systems are effectively black boxes (though implementations of state-of-the-art techniques do exist), and address devices hijacked with short-code voting.
• "Political Participation and Security Risks of Internet Voting in Ukraine under Martial Law" (2026) — DOI: 10.31861/mediaforum.2026.18.98-119 — Specifies remote voter identification issues as a major threat, along with the remote voting demand demographic (soldiers, internally displaced persons, overseas citizens).
• "Review Your Choices: When Confirmation Pages Break Ballot Secrecy in Online Elections" (2022) — DOI: 10.1007/978-3-031-15911-4_3 — Infers choices using only network data length, with 100% accuracy in some cases. Coordinated disclosure and deployment of mitigations.
• "Re-voting Under Surveillance: National eID Transaction Logs as a Threat to Coercion Resistance in Estonian Internet Voting" (2025) — DOI: 10.1007/978-3-032-05036-6_12 — eID logs reveal the number and time of revotes, breaking coercion resistance. Responsible disclosure completed.
• "Norwegian internet voting protocol revisited" (2016) — DOI: 10.1002/sec.1678 — Collapse of privacy when the ballot box and receipt generator collude. Holders of postal + SMS codes can prove their vote afterward.
• "Security Seals on Voting Machines" (2011) — DOI: 10.1145/2019599.2019603 — Tamper-evident seals can usually be easily bypassed, and their effectiveness depends on the application and inspection protocol.
• "Risk-Limiting Audits by Stratified Union-Intersection Tests of Elections (SUITE)" (2018) — arXiv:1809.04235 (preprint) — Definition of statistical guarantees for RLAs and stratified sample auditing.
• "Implementing Risk-Limiting Post-Election Audits in California" (2009) — arXiv:0905.4691 (preprint) — Pilot audit of 4 elections across 3 counties in 2008. The current approach is too complex for large-scale routine use.
• "PROVIDENCE: a Flexible Round-by-Round Risk-Limiting Audit" (2022) — arXiv:2210.08717 (preprint) — Actually used to audit the tally of the Rhode Island Board of Elections in the 2021 election.
• "Trust in the Count: Improving Voter Confidence with Post-election Audits" (2024) — DOI: 10.1093/poq/nfae029 — How an audit is conducted matters more to voter evaluation than what it finds.
• "Secure and Verifiable Electronic Voting in Practice: the use of vVote in the Victorian State Election" (2015) — arXiv:1504.07098 (preprint) — The first deployment of polling-place end-to-end verifiability in a state statutory election. The first instance of a voter with visual impairments casting a verifiable, completely secret ballot, numbering 1,121 votes.
• "Publicly Understandable Electronic Voting" (2026) — arXiv:2603.21833 (preprint) — 'Software-free verification', starting from the criticism that the cryptographic reliance of end-to-end verifiability becomes a new black box.
• "TRIP: Coercion-resistant Registration for E-Voting with Verifiability and Usability in Votegral" (2022) — arXiv:2202.06692 (preprint) — Printing real and fake credentials via a physical polling booth kiosk. Issuance delay up to 19.7 seconds.
• "On the Necessity of Pre-agreed Secrets for Thwarting Last-minute Coercion" (2026) — arXiv:2604.00188 (preprint) — Loki's two vulnerabilities and the dilemma that "last-minute coercion cannot be prevented without pre-agreed secrets."
• "Expanding the Toolbox: Coercion and Vote-Selling at Vote-Casting Revisited" (2024) — DOI: 10.1007/978-3-031-72244-8_9 — Coercion attacks using blockchain, time-lock cryptography, and trusted hardware. Calls for rewriting the definitions of coercion and no-receipts.
• "E-Vote Your Conscience: Perceptions of Coercion and Vote Buying, and the Usability of Fake Credentials in Online Voting" (2024) — arXiv:2404.12075 (preprint) — 96% comprehension, 53% willingness to use, 10% voting errors, SUS 70.4.
• "What Voters are Asked to Verify Affects Ballot Verification" (2018) — DOI: 10.2139/ssrn.3292208 (preprint) — Most voters do not verify summary cards, and even when they review them, they cannot remember their own vote from a few minutes prior.
• "Towards end-to-end verifiable online voting: adding verifiability to established voting systems" (2019) — arXiv:1912.00288 (preprint) — Adding a verification layer to operational commercial systems, with usability evaluated positively across two pilots.
• "E-voting: Lessons from Estonia" (2013) — DOI: 10.1007/s12290-013-0261-7 — Operating since 2005. About 44 minutes waiting time at polling stations versus 6 minutes for electronic voting.
• "Could Internet Voting Halt Declining Electoral Turnout? New Evidence That E-Voting Is Habit Forming" (2017) — DOI: 10.1002/poi3.160 — Turnout increases were minimal, but e-voting is very 'sticky' (five national elections from 2009 to 2015).
• "Trust in electronic voting: the Estonian case" (2024) — DOI: 10.21564/2414-990x.166.310197 — Points out the absence of independent observers and flaws in the appeal procedure. Calls for end-to-end verification and high-quality audits at every stage; complete replacement is premature.
• "Electoral Integrity Without Disenfranchisement" (2026) — DOI: 10.2139/ssrn.7159338 (preprint) — A redesign proposal to make vote-buying transactions unverifiable to the buyer rather than filtering out voters.
• "Hacking Voters' Trust in Democracy: Panel Evidence on Safeguarding Confidence in Election Integrity" (2026) — DOI: 10.1007/s11109-026-10147-6 — Cyberattack reports undermine perceptions of election integrity across party lines. The same goes for attacks unrelated to elections. Pre-bunking videos maintain trust.