Switzerland’s e-voting trials: suspension, redesign, and the 2026 Basel-Stadt incident

Switzerland has trialed internet voting since 2004: by the Federal Chancellery’s count, 15 cantons have offered it to a limited part of their electorate in more than 300 trials.⁠¹ In 2019, researchers using Swiss Post’s publicly released source code found that its system could produce “proofs” of a correct result even when votes had been manipulated, which led to the system’s withdrawal and a federal redesign;⁠^2,3 trials resumed in 2023 under stricter rules, and in March 2026 about 2,048 e-votes in Basel-Stadt could not be counted because the decryption key on a USB stick could not be used.⁠^7,8

In plain words

Switzerland votes a lot, often four times a year on national questions, and many Swiss citizens live abroad. Voting by mail from far away can be slow. So for twenty years, some Swiss cantons (the country’s states) have let a limited group of voters cast ballots over the internet.

In 2019, Swiss Post had a new e-voting system it wanted approved for wide use. Swiss law required that the code be published so anyone could study it. Independent researchers did, and found serious mistakes in the math meant to prove that votes were counted correctly. A dishonest insider could have changed votes and still produced a “proof” that looked fine. Swiss Post pulled the system and the government stopped and rethought the program.

The rules were rebuilt: only systems whose results can be fully checked, published code, a reward program for people who find bugs, and regular independent audits. Trials restarted in 2023. Then, in March 2026, one canton could not unlock about two thousand online votes, because of a problem with the USB sticks that held the secret key. The votes could not be counted. The canton paused its trials.

Defined terms

• Complete verifiability. The Swiss requirement, as the researchers summarize the ordinance, that individual verifiability hold as long as at least one server side component is honest, combined with universal verifiability that the result is correct under the same condition.⁠³ The Federal Chancellery states that only systems with complete verifiability are authorized in Switzerland.⁠⁴
• Individual verifiability. Each voter can check that the vote was recorded as intended. In the Swiss Post design, each voter receives by mail a code sheet with a secret “choice code” for each option and checks that the codes returned after voting match.⁠³
• Universal verifiability. Anyone (or designated auditors) can check that the published result follows from the recorded encrypted votes, usually through zero knowledge proofs of correct mixing and decryption.
• Mix net. A chain of servers that shuffle and re-encrypt ballots so they can be decrypted without linking them to voters; each server must prove it shuffled honestly.⁠³
• Zero knowledge proof (ZKP). A cryptographic proof that a statement is true without revealing why, such as proving a shuffle preserved every vote without revealing the permutation.
• Basic license (Swiss official English: “basic licence”). A Federal Council authorization allowing a canton to run trials for a limited section of its electorate for a set period; cantons must also be authorized for each federal vote.⁠^6,8

What happened

Switzerland’s early trials used systems developed by cantons and vendors. Geneva’s CHvote platform, developed in 2003, served Geneva and was used by Bern, Aargau, and Lucerne from 2010.⁠⁹ Geneva announced in November 2018 that it would shelve CHvote for cost reasons and later shut it down earlier than its planned February 2020 end, leaving those cantons’ voters without e-voting in the October 2019 federal elections.⁠⁹

That left Swiss Post’s system. Its version 1.0 had been used since 2016 in four cantons and certified for up to 50 percent of voters; version 2.1 was proposed in 2018 for certification for up to 100 percent of voters in participating cantons.⁠³ The Federal Chancellery announced a public intrusion test of the new system for February 25 to March 24, 2019, saying “interested hackers from all over the world are welcome to attack the system”.⁠⁵

On March 12, 2019, the Chancellery announced that “researchers have uncovered a significant flaw in the new e-voting system of Swiss Post,” in the part offering universal verifiability, and said the flaw did not affect the system then in use in four cantons.⁠² On March 29, it announced that Swiss Post’s system “will not be available for the vote on 19 May” and that it would review licensing and certification.⁠¹⁰ On June 26, 2019, the Federal Council “decided to provisionally forgo introducing electronic voting into regular operation,” noting that most consultation respondents supported e-voting but “considered its introduction into regular operation to be premature”.⁠¹¹

On December 18, 2020, the Federal Council launched a redesign built on “more precise security requirements,” “increased transparency,” “closer cooperation with independent experts,” and “effective auditing on behalf of the Confederation”.⁠¹² A revised Ordinance on Political Rights and a totally revised Federal Chancellery Ordinance on Electronic Voting took effect on July 1, 2022.⁠¹³ On March 3, 2023, the Federal Council granted basic licenses to Basel-Stadt, St. Gallen, and Thurgau, and the Chancellery authorized e-voting for the June 18, 2023 vote.⁠⁶ Those cantons used it in the October 22, 2023 National Council elections, and Graubünden relaunched trials in March 2024.⁠¹⁴

In June 2025, the Federal Council renewed basic licenses for four cantons through the June 6, 2027 vote; for the September 28, 2025 vote, the cantons sought authorization for about 159,500 voters, 2.84 percent of all Swiss voters.⁠¹⁴ In 2026 the Federal Council approved e-voting for Lucerne (June) and Neuchâtel (September).⁠^1,15

How it worked

The Swiss Post sVote 2.1 design examined in 2019 had several parts: a trusted print office that generates and mails code sheets, a voting client “trusted for privacy only,” a voting server, and server side control components that compute return codes and then mix and decrypt the votes.⁠³ The voter enters a vote on an untrusted device; the client sends an encrypted vote and partial choice codes with ZKPs that they are consistent; the servers compute choice codes and return them; the voter compares them with the paper code sheet. After voting closes, the votes are mixed and decrypted with proofs of correct mixing and decryption.⁠³

The design’s goal was that no single dishonest server component could alter votes undetected. The researchers note that the Swiss notion of universal verifiability is weaker than the traditional one: it guarantees correctness “as long as at least one server-side component is honest,” whereas the traditional notion holds even if all server components are dishonest.⁠³

Swiss rules made outside scrutiny a legal requirement. The researchers quote Ordinance 161.116 as requiring that source code “must be easily obtainable, free of charge, on the internet” and that “anyone is entitled to examine, modify, compile and execute the source code for ideational purposes, and to write and publish studies thereon”.⁠³ Under the rebuilt regime, the Federal Chancellery states that source code and documentation of completely verifiable systems “must be published,” that a bug bounty program runs on the disclosed code and documentation, and that independent organizations regularly examine the systems and their operation.⁠⁴ Licenses limit the electorate: in 2025, for example, St. Gallen and Thurgau were capped at 30 percent of the cantonal electorate, and Basel-Stadt offered e-voting to voters abroad and voters with disabilities who register.⁠¹⁴

Evidence: measured findings

The core 2019 evidence is the peer-reviewed paper by Thomas Haines, Sarah Jamie Lewis, Olivier Pereira, and Vanessa Teague, “How not to prove your election outcome,” presented at the 2020 IEEE Symposium on Security and Privacy.⁠³ Its abstract states: “We show failures in both individual verifiability and universal verifiability (as defined in Swiss Federal Ordinance 161.116), based on mistaken implementations of cryptographic components. These failures allow for the construction of ‘proofs’ of an accurate election outcome that pass verification though the votes have been manipulated”.⁠³

The authors group the failures into four generalizable errors:⁠³

1. Statically secure proofs used adaptively. Proofs secure only when the statement is fixed in advance were used where a cheating prover could choose the statement afterward.
2. Proven facts not sufficient. The client’s proofs did not, taken together, establish that a vote was well formed.
3. Collusion. One of five code computing authorities, acting alone with a cheating client, could break verifiability.
4. Unverifiable parameters. There was no evidence of how group parameters were generated, so a party could hold a trapdoor that violates both individual and universal verifiability.

They report that “every one of these errors allows a successful vote manipulation that passes verification,” that one weakness was also present in sVote 1.0, already used in binding elections, and that another was present in a system used in a New South Wales state election.⁠³ They published cheating proof transcripts so others could confirm the proofs pass verification.⁠³ Because of these findings, they write, “no version of sVote was used during the May 2019 Swiss referendum or the Federal elections of October 2019”.⁠³

The March 2026 incident produced a second set of official numbers. On March 5, Basel-Stadt told the Chancellery that electronically cast votes might not be decryptable; the problem was traced to “an external hardware component (portable storage medium)” holding the decryption key, and the canton urged about 10,300 eligible voters abroad and 30 voters with disabilities to vote by post or in person if possible.⁠⁷ Afterward, the Chancellery reported that about 2,048 people had voted electronically and those votes could not be counted; it judged the provisional results clear enough that they could not have changed any outcome, but said not counting them violated those voters’ political rights.⁠⁸ It attributed the incident to “irregularities in handling PIN-protected USB sticks,” said human error could not be ruled out, found nothing pointing to a fault in Swiss Post’s system, noted that the public prosecutor had opened criminal proceedings on an initial suspicion of a possible offense, and required all e-voting cantons to review how they protect decryption keys.⁠⁸

Failures and criticisms

Switzerland’s 2019 failure was a certification failure as well as a vendor failure. Swiss Post’s version 1.0 had been certified and used in binding votes while carrying a weakness later found by outside researchers.⁠³ The researchers argue that “if a cryptographic protocol is designed or implemented in secret, if it comes with no convincing proof of the soundness of its verification process, and if no opportunity for independent scrutiny is given, then it is unlikely that it offers the security properties it advertises”.⁠³

The terms of scrutiny drew criticism. The code was released under a non-disclosure agreement requiring that no vulnerability be published within 45 days of the last communication with the owners; the researchers say the code also circulated freely online, which is how they obtained it.⁠³ Critics of such terms argue they discourage the open analysis the law intended; the Chancellery’s current approach emphasizes publication and a bug bounty.⁠⁴

The Chancellery’s first statement said the flaw “does not allow the system to be intruded”,⁠² framing it narrowly, while the researchers stressed that it allowed undetectable manipulation by insiders.⁠³ Both can be true: the flaw was not an outside break in, but it defeated the property the system existed to provide.

The 2026 incident shows a different failure class. The cryptography and servers worked; a key custody procedure did not.⁠⁸ In a paper election, an analogous failure would be a locked ballot box whose key was lost; with encryption, there is no locksmith.

Competing interpretations

The Swiss model works because it catches failures. Supporters point out that the legal requirement to publish code is what exposed the 2019 flaws,⁠³ that the government stopped rather than pressing ahead,⁠¹¹ and that the rebuilt program adds independent examinations, a bug bounty, and electorate caps.⁠^4,14 In 2025 the cantons and the Chancellery said they were satisfied with the results so far.⁠¹⁴ In 2026 the Chancellery found no reason to question the other cantons’ licenses.⁠⁸

The Swiss record shows remote voting remains fragile. Skeptics note that a certified system carried a weakness into binding votes,⁠³ that the verifiability standard depends on at least one honest server component,⁠³ and that in 2026 more than two thousand voters lost their votes to a USB stick problem.⁠⁸ They also note what verifiability cannot address: coercion in an unsupervised setting.

Access as the overriding value. For many Swiss voters abroad, postal voting is slow and unreliable, which is why trials target them.⁠¹⁴ One reading is that a small, capped, carefully audited channel for people otherwise excluded is a reasonable trade even with residual risk; another is that the channel should not exist until risks are lower. Both are value judgments the Swiss political process continues to weigh.

Lessons

These are interpretations drawn from the record.

1. Openness is a security control, not a courtesy. Mandatory code publication found what certification missed.⁠³
2. “Verifiable” is a property to test. Sophisticated proofs can be implemented wrongly and still produce outputs that look like proofs.⁠³ See audits and verification.
3. Scale limits buy time. Caps on the share of the electorate limited exposure while flaws were found.⁠^3,14
4. Availability is part of integrity. A vote that cannot be decrypted is a vote lost.⁠⁸ Key ceremonies, backups, and custody need the same scrutiny as code; see hardware.
5. Stopping is a legitimate outcome. The 2019 pause and the 2026 Basel-Stadt suspension until December 31, 2026⁠⁸ show institutions treating halts as normal risk management.

Application to this framework

Switzerland’s program speaks to the My Vote Forward framework’s requirement that any future official voting component rest on evidence the public can check, not on an operator’s assurance (see security). The framework’s governance principles call for published methods and independent review; the Swiss law turns that principle into an enforceable duty.

The case also illustrates the framework’s distinction between the private vote and public, voluntary expression. Swiss verifiability is designed to let a voter check a vote without being able to show it to others, which is the property the framework’s private voting chapter asks for. And the 2026 incident informs the framework’s published releases approach: any key that protects data is also a single point of failure that needs a documented recovery plan.

Unresolved questions

• What did Basel-Stadt’s external analysis and the criminal investigation conclude about the March 2026 key failure? Neither had been reported in the sources reviewed here.
• Is a trust model that requires at least one honest server component adequate for national elections, or only for capped trials?
• How should a jurisdiction recover when encrypted ballots cannot be decrypted: rerun, accept the loss, or build in key escrow that itself creates risk?
• Can remote verifiable voting address coercion as well as it addresses manipulation by servers?
• What would justify raising electorate caps, and who should decide?