Author: rakerman

Internet voting must be about public evidence not belief

Internet voting, and indeed any kind of trusted election must be about public evidence, not belief.

If we wanted to conduct elections based on belief, we’d just take all the ballots into a secret room and say “trust us, we believe we have all the right counting and integrity in place”, and then produce the final count of the votes basically out of nowhere.

If we did that with paper ballots people would be incredibly suspicious.  Who did the counting?  How can we be sure the ballots were honestly counted?  Where was the oversight?  Where are the ballots to provide the evidence?  Can we even trust the ballots now that they have been held in secret?  What if they were changed?

This seemingly-ridiculous scenario is actually a pretty accurate description of where Canada is now with Internet voting.

A typical “debate” scenario has a Chief Electoral Officer or city councillor or city staffer on one side, and a computer scientist on the other.  Not only is this a totally artificial “balance” of views, the main issue becomes assertions of belief without evidence, on both sides.

The electoral officer says believe us, we have all the necessary measures in place to make Internet voting trustworthy.  The computer scientist says they believe there are possible attacks.  And that’s it.  You’re left to try to decide which belief to believe.

Fundamentally elections are not supposed to work like this.  Elections are not about trust and belief, they’re about evidence.

Maybe after having the anonymous paper ballot for so long we’ve forgotten that it was designed to provide public evidence, it’s not just a haphazard system we ended up with.

So Internet voting must provide public evidence, but it doesn’t.  Internet voting in Canada should provide public source code, but it doesn’t.  Internet voting in Canada should provide a public opportunity to conduct realistic attacks on the real system, or a very close model of the real system, but it doesn’t.  Internet voting in Canada in fact produces zero public evidence.  In fact, both the provision of public source code and public attacks on the real system are illegal, the former because of intellectual property law and the latter because of cybersecurity law.  Which is why the computer scientist can only say “believe me, there are potential attacks” rather than actually demonstrating real attacks.

So for Internet voting, you now have to entirely transfer your trust to the election organisation, but actually it is worse than that, because with the third-party vendor model of Internet voting that Canada uses actually you’re entirely transferring your trust to the third-party, for-profit vendor.

What security tests are conducted on the vendor?  Sorry, that’s a secret.

What security measures are taken by the elections organisation?  Sorry, that’s a secret.

What security measures are in the code and the servers and the network the vendor provides?  Sorry, that’s a secret.

To be clear, I have a high degree of confidence that Canada’s public election organisations are doing their job with all necessary diligence and expertise.  But my confidence is irrelevant.  Confidence is not how you run elections, evidence is.  And in the transition to computer vote counting and Internet voting, we have totally changed our trust model without any meaningful public discussion (as I have mentioned before specifically about computer vote counting).

Maybe from now until the end of time, our public election organisations and their private vendors with secret code and secret testing will conduct themselves perfectly.

But this seems unlikely given human history plus the fact that every single time voting code is made available for inspection or opened to public attack, the code is shown to be insecure.  This ranges from Washington, DC in 2010 as documented by J. Alex Halderman, to Switzerland in 2019.

There are very good computer-theoretic reasons that you can’t trust Internet voting even if the code is found to be secure, including under real-world attack; there is as yet no solution for secure Internet voting.  But it is perfectly reasonable to experiment in low-risk, small-turnout situations.

An actual experiment would place Internet voting in the same space of public evidence as paper ballots.  Which means that Canada would need standards, public code and public testing.  You may be shocked to find out that unlike pretty much everything from your municipal water supply to any product you may buy, Canada has no, none, zero standards for Internet voting.  No mandatory requirements.  No mandatory testing.  No nothing.  Internet voting typically shows up as a single line about “electronic voting” in an alternative voting methods law or bylaw.  That’s it.

The absolute critical first step to bringing public evidence back to elections in the Internet voting era is to have some very basic foundational standards and requirements, starting for example with the Swiss model that requires both public source code and public security testing.

In the absence of bringing public evidence to the conversation about Internet voting, we’re just going to have year after year of the same pointless back and forth about election beliefs, a conversation that can never be resolved because there’s no actual evidence to draw conclusions from.



Estonian Parliamentary Elections 2019 – ODIHR Election Expert Team Final Report – Internet Voting

The Office for Democratic Institutions and Human Rights (ODIHR) is a division of the Organization for Security and Co-operation in Europe.  The ODIHR has produced a report on the 3 March 2019 Estonian Parliamentary Elections.

ODIHR Election Expert Team Final Report – Estonia – Parliamentary Elections 3 March 2019 (PDF)

The ODIHR reviews a wide range of election conduct against international standards.  I will only extract selected parts of their report from section VII. Internet voting.  Numerous issues were identified.

In extracts below, EET = Election Expert Team and SEO = Estonian State Electoral Office.

Internal Attacks

the detection and prevention of internal attacks has been largely omitted. A review of operational and technical frameworks by the ODIHR EET indicates that an internal attacker with privileged access to digital ballots could break the vote secrecy of any voter who published an image of the QR code online, even after the expiry of the code’s validity. This contradicts national legislation and international standards pertaining to vote secrecy.21

RECOMMENDATION: The SEO could develop strategies to mitigate the risk of internal attacks, conduct third-party risk assessments, and publish any findings and audit reports well ahead of the next elections.

21 See Article 1(2) of the Election Act. Paragraph 7.4 of the OSCE Copenhagen Document requires that votes are cast by secret ballot or by equivalent free voting procedure. Paragraph 19 of the Council of Europe Committee of Ministers Recommendation CM/Rec(2017)5 on standards for e-voting requires that “E-voting shall be organized in such a way as to ensure that the secrecy of the vote is respected at all stages of the voting procedure”.

above from page 8 of the report

Software Errors May Cause Election Errors

The Internet voting system is not software independent, meaning that software errors in its components, such as the key generation system or the processor, may cause undetected errors in the election results. Considering publicly available records the system has undergone quality control activities but, contrary to international good practice, no reports were published on the SEO’s website, while updates to the source code were made as recently as three days before election day and well after Internet voting commenced.22

In addition, a limited source code review of the system by the ODIHR EET indicated issues regarding the treatment of concurrency, error handling, and error reporting.

RECOMMENDATION: The SEO could integrate quality assurance activities into the maintenance schedule of the voting solution and publish the security rationale and all quality assurance results, including design review, security analysis, and penetration testing results.

22 Paragraph 42 of the Recommendation CM/Rec(2017)5 on standards for e-voting states that “Before any e-election takes place, the electoral management body shall satisfy itself that the e-voting system is genuine and operates correctly.”

above from page 8 of report

External Auditors Did Not Audit All Operations

A team of external auditors was dispatched to assist the SEO with establishing vote secrecy during the computation of preliminary Internet voting results and the integrity of final Internet voting results by verifying the correctness of the cryptographic shuffle and decryption proofs. The team did not audit other critical operations, most notably the correct transmission of the final aggregation of the decrypted Internet votes.23

RECOMMENDATION: The SEO could strengthen its auditing process by developing a complete strategy and requiring auditors to implement critical auditing tools independently and from scratch.

23 Software independence requires that other operations are also independently audited, such as digital signature checking of all e-votes, removal of all duplicate and other ineligible votes from the digital ballot box, revocation, and anonymization. Paragraph 39 of the Recommendation CM/Rec(2017)5 on standards for e-voting states that “the audit system shall be open and comprehensive, and actively report on potential issues and threats.”

above from page 9

Technical Specifications Need Improvement

some key properties are not precisely formulated and left open to interpretation by the SEO and the vendor tasked to implement the Internet voting system, including minimal acceptable levels of cryptographic strength, and accountability and verifiability requirements. This may negatively impact the system’s overall performance and future innovation. The specifications also lack information about timelines and milestones for software development and deployment, and quality assurance.25

RECOMMENDATION: The technological specifications accompanying the legal framework could define acceptable voting systems in more general terms, but include additional requirements related to cryptographic strength, quality assurance, software development and deployment, as well as accountability and verifiability.

25 The Supreme Court considered two post-election appeals against NEC decisions related to Internet voting. While appeals were rejected, the Court recognized the need for more clear procedures and called for a legal clarification of rules on the implementation of Internet voting, in particular regarding counting and mixing of electronic ballots.

above from page 9

Internet voting at the national level

National-level Internet voting

  • Norway discontinued Internet voting trials in 2014.
  • Australia recommended against Internet voting in 2014.
  • Canada recommended against Internet voting in 2016.  The 2019 national Parliamentary election will have hand-marked paper ballots, counted by hand.
  • Finland studied and recommended against Internet voting in 2017.
  • Lithuania has decided not to proceed with Internet voting in 2019.
  • Switzerland has decided to redesign Internet voting trials, rather than making Internet voting a standard option.  However the Swiss Post system has been “temporarily suspended” after critical errors were found in the source code, and the use of the Geneva system has also been suspended pending a review.

UPDATE 2019-07-07: Swiss Post has made a confusing press release, basically to say that it will continue with its new system and discontinue its old one.  The “new system” is the one that had the public testing.  The public testing in which, through access to the source code outside of the restrictive agreement, three serious flaws in the system were found.

Swiss Post has decided to pool its strengths in the e-voting sector and work solely on the new system with universal verifiability. It plans to make the system available to the cantons for trial operation from 2020. Swiss Post will no longer offer the system that was previously in use.


Estonia continues to be the only country in the entire world that has national-level Internet voting for all voters (during the advance voting period).  And it has numerous issues with procedures and specifications, as well as low and declining turnout.

Although not directly about Internet voting, also note:

Internet voting in Norway

Norway conducted trials of Internet voting in 2011 and 2013.

Internet voting was discontinued after the trials found no improvement in turnout (including no increase in youth turnout), combined with security concerns.

An archive of reports in Norwegian and English is available: The e-vote trial.

Here are some highlights of the reports:

Evaluation of the e-voting trial in 2011 – English summary of Institutt for Samfunnsforskning (ISF) report

we find no evidence that groups of voters have been mobilized to take part in the election as a result of internet voting.

The analyses, in sum, indicate that the trial did not have an effect on voter turnout.

young voters prefer to walk to the polling station on Election Day. They defined traditional voting as a symbolic and ceremonial act that indicates adultness.

Evaluation of the e-voting trial in 2013 (PDF) – English text begins on p 135 (p 137 in PDF)

In line with previous research, our findings indicate that the trial with internet voting does not lead to increased turnout in elections.

The government announced in 2014 that Internet voting trials would be discontinued.

June 25, 2014 – Internet voting pilot to be discontinued

As there is no broad political desire to introduce internet voting, the Government has concluded that it will would be inappropriate to spend time and money on further pilot projects.

The Institute for Social Research evaluated the pilot project in 2013… The report shows that the voters have limited knowledge about the security mechanisms in the system.

“This shows how important it is that elections are conducted at polling stations where election officials make sure that the principle of free and fair elections and the secrecy of the vote is respected,” says [Minister of Local Government and Modernisation Jan Tore] Sanner.

In Norwegian – Ikke flere forsøk med stemmegivning over Internett

The BBC reported this as E-voting experiments end in Norway amid security fears.

As part of the project, in 2009 there was a report on security.  It notes the added risks from remote voting.

The system is no longer by necessity confined to the local polling station; conceivably it is accessible world-wide, thus increasing the potential number of attackers and attack vectors dramatically.

Also as part of the project, in 2012 there was 196-page report International Experience with E-Voting [with a focus on Internet Voting] (PDF).

Online voting and turnout in the 2019 European Parliamentary elections

I am grateful to Estonia for publishing detailed turnout statistics online on its official government elections website, in Estonian and English.

The 2019 European Parliament elections have completed.  Estonia was the only country to offer online voting.  There were seven days of online voting available during the advance voting period in Estonia, from May 16 to May 22 inclusive.

The turnout (percentage of eligible voters) for online voting was 17.6%.

Over 80% of eligible Estonian voters chose not to vote online.

The total turnout was 37.6%.

The majority of Estonian voters chose to vote on paper.

Estonian turnout increased 1.1%.  But provisional overall European turnout increased over 8%.

Estonian turnout was 37.6%.  But provisional overall European turnout was 50.97%.

Estonia’s neighbour Finland doesn’t vote online.  Provisional turnout in Finland was 40.7%.

Estonia’s neighbour Lithuania doesn’t vote online.  Provisional turnout in Lithuania was 53.08%.

In fact, with no other country in this election permitting online voting, turnout was higher in 19 of the 27 other countries that voted.

European Parliament 2019 provisional turnout
Estonia had lower turnout than 19 of the 27 other countries, lower turnout than overall in Europe, and a lower increase in turnout than overall in Europe.

Online voting doesn’t put Estonia at the top of the pack for turnout.  In fact Estonia was in the bottom third of nations for turnout in the 2019 European Parliamentary elections.

So I don’t know how one could continue to assert that online voting is any kind of solution to increasing turnout.

UPDATE 2019-06-01: I have made a Google Docs spreadsheet of the European Parliamentary elections turnout data, if you want to look at the numbers yourself.  Note that this shows total turnout; as indicated above Estonia votes online and on paper, with the majority voting on paper.  END UPDATE

Data sources:

March 5, 2019  Internet voting doesn’t increase turnout in Estonian elections

online voting is destructive modernization

Is Internet voting beneficial modernization, or destructive modernization?

Let’s use an analogy to urban life and urban planning.  I am all for beneficial modernization.

Philadelphia, like all cities before the 20th century, was a hotbed of infectious disease.  Living in a city meant enduring wave after wave of epidemics, and those who could stayed outside the city until the disease burned itself out.

The departure from Quincy took place in the first week of October [1797], but the Adamses were not to reach the capital for more than a month.  Yellow fever again raged in Philadelphia, as they learned en route, so it was necessary to stop and wait at East Chester…

from John Adams by David McCullough

So the sanitation and medical advances that made cities livable were definitely beneficial modernization.

However, modernization can also be destructive.  After the second World War, a wave of “urban renewal” swept over cities, a wave almost as destructive as any epidemic.  The messy diverse urban reality of the city was to be swept away, there was to be a clean slate, everything would be modern and efficient and the car would be unhindered in its rapid progress through the city.

Philadelphia, why Philadelphia was a blank canvas upon which men could draw their geometric visions, without any of the inconvenient reality of poor people, minorities, women, children or the elderly.

Form Design and the City
above from Form, Design and the City (1962)

The city would be clean, modern, efficient.  Le Corbusier looked at Paris and instead of seeing beauty, he saw something he needed to fix.  Here’s his idea for Paris, the Plan Voisin
Plan Voisin model
above by SiefkinDR [CC BY-SA 4.0], via Wikimedia Commons

Robert Moses famously pushed highways throughout urban New York, and as in many parts of the US and Canada, the highways often went through poor neighbourhoods, wiping them out or splitting them permanently in two.  Cars would be convenient, cars would go fast. He was only stopped when he wanted the Lower Manhattan Expressway, stopped in part by Jane Jacobs.

Jane Jacobs wrote about the reality of human behaviour in The Death and Life of Great American Cities

This last point, that the sight of people attracts still other people, is something that city planners and city architectural designers seem to find incomprehensible.  They seem to operate on the premise that city people seek the sight of emptiness, obvious order and quiet.

This is part of her explaining her concept of security through “eyes on the street”.

When we look at online voting–you remember this is about online voting, right?–what we see is exactly those same assumptions, that we should remove people from the equation, that we should make everything orderly and “efficient”.

But voting is not a generic service like paying a parking ticket.  Voting is a complex transaction where you need the paradoxical combination of total privacy and total observation.  Voting is also a social contract where a single vote is transformed into trust in the results of an entire national election.  Voting is about being part of your community.  Voting does not need false notions of convenience and speed.  In order to transmute the individual vote into trust in the entire system, voting needs to move at a human pace under watchful human eyes.  Voting is, in other words, a very human activity.  And so when you move it online, when you drive that highway through the messy line of people waiting to vote, you’re engaged in destructive modernization.

One of the things that “urban renewal” and urban highways did was make life worse for the already underprivileged, and better for the already powerful.  In other words, the white middle-class men who did the planning benefited other white middle-class men.

If we apply this same test to online voting, we find exactly the same thing: the people that actually vote online are the ones who would vote anyway: older male voters.

To put it in the words of What drives fidelity to internet voting? Evidence from the roll–out of internet voting in Switzerland

  • Lower age cohorts are the least likely to remain faithful to internet voting.
  • Senior voters are more likely to remain faithful to internet voting.
  • Gender also has an effect, with women less likely to remain faithful to internet voting.

Or in the rather clearer words of the 2014 BC Independent Panel on Internet Voting

research suggests that Internet voting does not generally cause nonvoters to vote. Instead, Internet voting is mostly used as a tool of convenience for individuals who have already decided to vote.

(See Online voting doesn’t increase turnout for more information on who actually votes online.)

So basically online voting removes the diverse participation and personal experience of casting your ballot on paper, it removes the “eyes on the street” of being able to see the process right in front of you (including the ability of scrutineers to watch the ballots being counted), and it benefits the already empowered people who would have voted anyway. It’s destructive modernization.

Swiss voting technology law sets the standard, in theory

Switzerland – Federal Chancellery Ordinance on Electronic Voting 161.116 of 13 December 2013 (Status as of 1 July 2018)

Key Concepts in Theory

  • the system must be independently evaluated (Article 7, item 1)
  • risk must be assessed (Article 3)
  • the system must be evaluated against detailed requirements (Article 2, section a, Article 4, Article 7, item 2 and item 3)
  • the source code must be made available (Article 7a and Article 7b)

Also notable is that the default maximum authorised participation in electronic voting is 30%.  From above 30% to 50% additional requirements apply, and above 50% even more requirements apply.

In Practice

Unfortunately in practice, for a 2019 public intrusion test, the conditions on both the general testing and the availability of source code were restrictive.

There was not in any sense either unrestricted public testing nor unrestricted publically available open source code.

(If you’ve heard that the tested voting system was withdrawn when serious security flaws were found, this is true, but discovery of these security flaws happened through access to the source code outside of the restrictive agreement.)

My Recommendations

The Swiss ordinance has model principles that should be adopted for evaluating online voting.  In particular independent public evaluation and availability of public source code are key (although keep in mind that source code availability doesn’t mean perfect confidence in the code that actually runs).

The Swiss law is however too complex, and it allowed the interpretation loopholes that led to restrictive terms of use in practice.

Therefore the model principles for evaluating online voting must also include clear language on unrestricted public testing and unrestricted public access to source code.

It’s also important that the independent testing include not just funded open hacking competitions (which are useful) but also direct funding to academic research groups.  The cryptography used in modern voting systems is extraordinarily complex; the academics who are expert in it don’t have free time and don’t work for free.

(Even with academics funded to study the voting system, be mindful that nation-state attackers have far more time and resources to devote to finding flaws in systems, as well as having arsenals of zero-day attacks they could choose to deploy during an election.)

Detailed Technical Language

Below are extracts of the technical language from the ordinance.

Voting System Must Meet Requirements

Art 2. … The authorisation for electronic voting in any individual ballot shall be granted provided the following requirements are met:

The system for electronic voting (the system) is implemented and operated so as to guarantee secure and trustworthy vote casting (Annex No 2 and 3).

There Must Be A Risk Assessment

Art 3. … By the means of a risk assessment, the canton must document in detailed and understandable terms that any security risks are within adequate limits. The assessment covers the following security objectives:

the accuracy of the result;
the protection of voting secrecy and non-disclosure of early provisional results;
the availability of functionalities;
the protection of personal information about voters;
the protection of voter information against manipulation;
the non-disclosure of evidence of vote casting behaviour.

Progressively Higher Requirements As Authorised Participation Increases

The ordinance takes an unusual approach which is to set progressively higher bars to increased availability of online voting. By default, the maximum percentage of the Swiss electorate allowed to use online voting is 30 percent (30%).

At 30% participation there is a minimum set of validation requirements

Art 7. 3If no more than 30 per cent of the cantonal electorate are to be authorised to participate in a trial and the system has the property of complete verifiability in terms of Article 5, the system and its operation must be examined in particular detail with regard to the following criteria:

cryptographic protocol (Annex No 5.1);
functionality (Annex No 5.2), whereby the examination may exclude the software in portals of authorities that are linked to a system;
security of infrastructure and operation (Annex No 5.3), whereby the examination may be limited to the infrastructure that registers the vote and creates the proof for the voter in accordance with Article 4 paragraph 2;
protection against attempts to infiltrate the infrastructure (Annex No 5.5);
control components (Annex No 5.4).2

To exceed 30%

Art 4.1If a system is to be authorised to cover more than 30 per cent of the cantonal electorate, the voters must be able to ascertain whether their vote has been manipulated or intercepted on the user platform or during transmission (individual verifiability, Annex No 4.1 and 4.2).

along with other conditions

Above 30% participation there are also different validation requirements

Art 7. 2If more than 30 per cent of the cantonal electorate are to be authorised to participate in a trial (Art. 4 and 5), the system and its operation must be examined in particular detail with regard to the following criteria:

cryptographic records (Annex No 5.1);
functionality (Annex No 5.2);
security of infrastructure and operation (Annex No 5.3);
protection against attempts to infiltrate the infrastructure (Annex No 5.5);
requirements for printing offices (Annex No 5.6);
when using a system has the property of complete verifiability in terms of Article 5: control components (Annex No 5.4).

To exceed 50%

Art 5.1If a system is to be authorised to cover more than 50 per cent of the cantonal electorate, it must be ensured that voters or the auditors are able, subject to compliance with voting secrecy, to identify any manipulation that leads to falsification of the result (complete verifiability, Annex No 4.3 and 4.4).

along with other conditions

Independent Assessment

Art. 7 Requirements for examinations

1 The cantons shall ensure that meeting the requirements is examined by independent agencies. The examination is made in particular if the system or its operation has been changed in such a way that meeting the requirements for authorisation could be called into question.

Publication of Source Code

Publication of source code is required, but it’s tangled in the level of authorised participation and in other attributes, so I will just include the entire section

Art. 7a1Publication of the source code

1 The source code for the system software must be made public.

2 Publication shall take place when the system has the property of complete verifiability in terms of Article 5, and:

following the examination in accordance with Article 7 paragraph 2 if more than 30 per cent of the cantonal electorate are to be authorised to participate in a trial;
following the examination in terms of Article 7 paragraph 3 if no more than 30 per cent of the cantonal electorate are to be authorised to participate in a trial.

3 There is no requirement to publish the source code of the following:

third-party components such as operating systems, databases, web and application servers, rights management systems, firewalls or routers, provided these are freely available and regularly updated;
portals of authorities that are linked to a system.

1 Inserted by No I of the FCh O of 30 May 2018, in force since 1 July 2018 (AS 2018 2279).

Art. 7b1Modalities for publishing the source code

1 The source code must be prepared and documented according to the best practices.

2 It must be easily obtainable, free of charge, on the internet.

3 The documentation on the system and its operation must explain the relevance of the individual components of the source code for the security of electronic voting. The documentation must be published along with the source code.

4 Anyone is entitled to examine, modify, compile and execute the source code for ideational purposes, and to write and publish studies thereon. The owner of the source code may permit its use for other purposes.

1 Inserted by No I of the FCh O of 30 May 2018, in force since 1 July 2018 (AS 2018 2279).

Official Versions

As English is not an official language of Switzerland, the annexes to the ordinance and explanations about the ordinance are available only in German, French and Italian.  The annexes provide additional technical detail and there was also an explanatory report produced in 2018 providing context about the need to publish the source code.

UPDATE 2019-05-24: Also see the E-voting home pages and policy pages for each language