So, it's my great pleasure to introduce Carsten Schurmann from IT University of Copenhagen,
who's going to talk about trust assumptions, which is a big deal in election integrity.
So, thank you.
Thank you so much, Matt, for this very nice introduction.
I should introduce myself. My name is Carsten Schurmann,
and I have been working with elections for the last ten years.
And I'm actually a computer scientist by training.
It sounds really funny here, but okay.
I'm a computer scientist by training, and I do actually very theoretical stuff.
There's programming languages and so on.
And then one day I decided I need a good example to show the world why this stuff is actually good.
And I started elections.
And since then, elections has, in my life, taken on a life on its own.
And in 2017, I've become an election observer with the Carter Center to Kenya
to help the Carter Center sort out what actually is the role of technology in the Kenyan election.
Since then, last year, I went to the Philippines with the Carter Center and back to Kenya.
I've been to Estonia, and I've been also to the United States in 2018.
Kind of look at your elections.
It's very funny because nobody knows how many jurisdictions there actually are in the United States.
It's anything between 8,500 and 10,500.
But not even the NASA knows about these kind of things.
So, when you're an election observer, you suddenly start thinking about,
how am I actually going to observe technology in elections?
How do I know that something is right?
We have heard from Amanda, you can later on do an audit in order to kind of verify certain things.
But if you were at DEF CON last year, you might have, or two years ago,
I gave a talk on these small numbers that you actually get in risk learning audits.
They don't necessarily generate trust.
They can actually kind of be used in disinformation as well.
So, my talk basically, you know, is about trust assumptions,
which I found a very useful tool to think about,
be able to talk about technology that is, by its very definition, unobservable,
to kind of have something to say about the quality of elections.
And I'm going to kind of share a little bit of some case studies,
I hope you find exciting, from my travels around the world,
of shit that can go wrong, okay?
And how you could have actually observed it if you just had thought about the trust assumptions ahead of time.
So, my first slide basically is about, it's very general.
It's my way of thinking about technology, okay?
So, it is very easy to kind of build a voting machine.
Every student, after two weeks of studies, can build a voting machine.
It's absolutely trivial.
What is difficult is to kind of make sure that the voting machine works,
not in the lab environment, but in the adversarial environment,
where people are trying to attack it, trying to change the vote, okay?
So, it's difficult, it's easy to get a system going with functionality,
more difficult to kind of make it secure.
But even security doesn't imply trust, right?
So, you have to kind of think about mechanisms of verifiability,
of accountability to kind of explain to people, this stuff works,
and the reported winner of the election is actually the right one.
A risk-limiting audit is a perfect example for verifiability.
And it is even harder, based on my last year's talk,
to build a system that's actually trustworthy.
There will be always people who do not trust.
It doesn't matter how much verifiability you have,
sometimes it's very difficult to kind of convince half of the country
that this is actually the right result.
Okay, based on this, this is the way how I think about it.
And, you know, when people actually kind of procure technology,
when ministries think about technologies in the United States and in Europe,
maybe the United States have taken great leaps in the last 15 years,
but in Europe, many people in the ministries are actually lawyers.
They are not technical people.
What they think about is like a vendor comes and explains to them the functionality,
they think it is fantastic.
We have a system that automatically counts votes.
This is what we should buy.
Whereas you really want a system that lives on this level.
Now, after kind of giving this stuff much, much thought,
I actually kind of also try to understand how disinformation works in this whole thing.
And it turns out that if you have a system which just has functionality,
you can easily spread disinformation attacks about like,
it is not secure, the Russians have hacked it, the Iran people have hacked it.
But, you know, next level up, if you have a system that is secure,
you know, it still kind of gives rise to some kind of information.
It's more difficult because it is secure.
So, people have arguments against that the system is secure or not.
But it could be misconfigured.
The ballots design don't match.
The system is perfectly secure, but it still does have mistakes
and still does not provide the right answers.
And even if you have a system that is verifiable,
where you actually really can check the evidence,
especially when you talk about Internet voting systems,
there could be disinformation and campaigns about,
yes, but half of your voters have been coerced.
I've seen all of these kind of arguments happening in all of the world.
And the ideas of these pyramids shows you that, you know,
the trust assumptions are actually kind of really important
because the trust assumptions tell you something about what you have to believe
in order to believe that the system actually works correctly.
If you have a system that is just functional, you have to trust that it is secure.
You have to trust the election commission to not screw it up.
You have to trust the voters.
You have to trust the DEF CON people not to hack it in the middle of the night.
So, these kind of things is like the further down you come,
the more trust assumptions you actually have.
And, you know, maybe study paradoxically, if you have full trust,
there are no more trust assumptions necessary in the whole thing.
Okay. So, this is the story.
And now I want to kind of go through my case studies
and show you how to derive actually some trust assumptions very carefully
and then compare them.
Okay.
Antrim country 2020, all of you or most of you probably have heard about the incident.
Okay. So, there were some issues with ballot designs.
Okay.
And so, concretely, so this is all taken from a report by Alex Haldeman
who analyzed this thing in very great detail.
And it was not under lock by any judge.
So, it's freely available.
And you can see, you know, here two ballots.
There's the original ballot design and there's an updated ballot design.
Okay.
And when you squeeze your... you see what the difference is between the two?
Yes.
Yeah, exactly.
So, here there are four bullets and here there are five bullets.
Okay.
So, this is like... now, a computer system is really, really important
if something is five or four or six or seven.
I mean, it's a difference.
Six is not equal to seven.
Right.
It's different.
Now, when you have this kind of an updated ballot design,
now, there's things must have gone wrong.
Okay.
Things went wrong.
Now, it's not an attack, but it's just a problem of the system design itself.
And I would argue, if you kind of look at like all of the stakeholders
that actually are involved in running an election, you know,
and there are many, are completely understandable, right?
What are actually their jobs?
What are they doing?
Who is responsible for what?
Okay.
So, what actually really happened here?
And this is from Alex's report.
Okay.
There was an incident.
Okay.
And, you know, things didn't sum up correctly because in a computer system,
essentially, when you have a table with five columns, right,
and you put data in, or with six columns, and you put data in with five columns,
then everything switches, you know, moves by one.
And so, it could actually be that for one party,
suddenly the totals are being added up for another party, right?
And how can this actually happen?
How can we prevent it?
Okay.
So, there's a couple of observations, updates to the ballot designs
after the initial designs had already been loaded onto the memory cards.
Okay.
So, it was not only that there was a change in the ballot designs,
but now suddenly we have two different kinds of memory cards,
ones that are correct and ones that are broken.
Okay.
Number two, updated and initial ballot designs in circulation.
So, that means that voters were actually sent first the one ballot design
and then said, I'm sorry, so the old one away.
Here is a better one.
Use that.
Okay.
So, you can, of course, there's one trust assumption popping out here immediately.
The trust assumption is that we trust the voter to use the right form to vote.
Is that a reasonable assumption?
Probably not.
Okay.
So, if you think about this at the design stage of the election or the electoral program,
you should kind of think about it.
There might be an issue here.
Let's fix this before it is too late.
Okay.
Number three, the county did not test loading scanners results into its election management system.
Now, that's problematic.
Why?
You will see this actually in the report that is outlined that there has to be a quality assurance test one time,
but if there's a change, it is optional to do another test because you have tested it already once.
This is like, here's a trust assumption going on underlying it, which also is not 100% right.
Okay.
Memory cards and EMS of the election management system use different versions of the ballot designs.
Okay.
So, then everything is kind of screwed up.
So, what are these kind of trust assumptions which I just tried to kind of get out?
So, county officials are trusted to update ballot designs and communicate the changes to the relevant people,
which they actually haven't done in time.
Okay.
There must be deadlines involved to actually stop these updates.
Okay.
The voters are trusted to vote on the correct ballot.
Okay.
Yeah.
Okay.
That's nice, but, you know, it's not clear if it's a good thing.
Logic and accuracy testing, I said it already, trusted to repeat LA testing if necessary, but you have to do it at least once.
Okay.
So, this is why, like, after the changes, no more tests were conducted.
The EMS maintainer is trusted to configure the memory cards and is trusted to import the polling data back into the back system.
Yeah.
And the scanner maintainer is trusted to configure scanners for all ballots designed in use.
Okay.
But, you know, can the technology actually handle this?
Can you kind of configure your scanners in such a way that when you kind of get different ballot designs and stores in different places,
or is it just on database?
How is the whole thing?
It might be impossible.
So, when you kind of come up with these assumptions, you can actually see, like, okay, trusted voting on a correct ballot is probably unreasonable.
Okay.
Why?
Because the voters are not all clever.
You know, they don't use necessarily the last one.
Maybe they threw the last one out.
They couldn't find anymore.
They didn't remember that they actually got two of those.
Then we have...
Okay.
Was one too fast here?
Incomplete.
Okay.
Trusted to configure the memory cards.
Okay.
So, yeah.
Why is it incomplete?
I already forgot.
I had actually a note here.
One second.
Okay.
I was afraid of it.
I might forget what exactly that means.
Let's go to the next one, the broken one.
Okay.
Trusted to update the ballot designs.
That just didn't work because the people actually didn't behave as we trusted them to behave.
Okay.
And there's inconsistent.
Yes.
I would call this inconsistent if they had the choice to actually do it once or several...
They can do it once or they can do it several times, but which ones are they going to take now?
So, it should be kind of cleared up, these kind of things.
And these kind of assumptions are...
Here's another one.
Okay.
No, that was it already.
Okay.
So, we have a way, basically, when you kind of get a technology, when you describe a process,
you don't just have to take about the technology itself.
You don't have to talk about the people itself.
But with these trust assumptions, you describe what is the interaction between the different things and kind of study them.
Okay.
So, let's move on to the Philippines.
Okay.
So, the Philippines actually also use ballot scanners as they use in the United States.
And here you can see them.
You know, you feed in your ballots.
There are some of these machines over there.
Not exactly that model.
And then, you know, there's actually after the machine has scanned a ballot, it produces this paper, this VB pad.
They call it actually VB pad, the verified paper audit trail.
Although they have the ballots actually in storage.
So, okay.
So, yeah.
What could go wrong?
So, these are the machines used.
A vendor uses them.
And the ballots, they look a little bit like this.
Okay.
So, this is a picture I took.
And there are 800 ballots for all polling stations.
And because there's a lot of things of like, you know, ballot stuffing and so on.
So, the vendor has decided that these ballots are marked with a little barcode here to actually work only with this one scanner in this particular precinct.
And not with no other.
Okay.
Which sounds... that's a very good thing.
However, what happens if the scanner is broken?
Then you can't actually go anywhere else to scan those things.
Right?
And there's something about privacy as well.
Okay.
And then, you know, as a computer scientist, of course, I have to look into the barcode.
And you can see there's actually a sequential number.
1, 2, 3, 4, 5, all the way to 800.
So, you can actually break the privacy of the voter by just kind of looking at it.
The other thing is also because there are 800 voters registered to one precinct.
There are exactly 800 ballots.
If you screw it up, you're not entitled to get another ballot.
Okay.
So, if you make a mark accidentally someplace else and the scanner says, oh, I'm sorry, I can't read it.
That was your only chance to vote.
Okay.
So, now, you have also this kind of machines in the United States.
So, I come from Denmark.
It's a slightly different kind of system.
It's all on paper and it's all very old-fashioned.
Okay.
It's all hand-counted.
It's old-fashioned.
It's also only one check on a piece of paper.
You can sort them.
You can weigh them and then you can actually come up with the result.
You don't have different races, which makes it so difficult to run American elections.
Okay.
And so, what happens is...
How do you preserve the privacy in the United States?
How do you preserve the privacy, the voter privacy when you vote?
Yes.
But you have like a little sleeve, right, where you put it in.
The privacy sleeve.
And then you go to the scanner.
Okay.
So, here's a little video I took.
Let's see if it actually...
How does it actually run?
Okay.
Let me see.
How does it run?
Okay.
So, this is a ballot.
A scanner is not 100% right.
So, they took it out of the sleeve.
Okay.
So, there's clearly a problem with the scanner.
There are like five people standing around.
Okay.
So, there's one taking a picture of the error message.
Then they press again.
So, then it comes back out.
Okay.
Then they have to turn it around just because you might not have seen all of the options that the voter has checked.
So, then it goes back in.
Okay.
Of course, it doesn't work either.
So, it'll come back out.
And they're actually forced to do it four times.
So, vote privacy, okay, is not really guaranteed with these optical scan machines.
So, I think also in the United States, there's an issue with that.
But nobody, for some reason, talks about this.
It's mostly...
The United States is mostly about integrity and not vote privacy.
But vote privacy is clearly as important.
Yes.
Okay.
So, they're still hoping.
But, of course, there's another disappointment.
It will come back out.
Yes.
So, in this kind of case, you know, what they actually do in the Philippines is they send their machines to get repaired.
They have a contract with the vendor.
And the vendor promises a work turnaround of so and so many hours before the machine comes back.
And the voter has the option to put their ballots into some ballot box.
And then the election official is going to push them into the scanner.
Now, that relies on a really, really important trust assumption.
Namely, that the voter trusts the election officials.
Which, unfortunately, is wrong.
Okay.
So, nobody, I have observed, is actually taking part of this option.
So, there was one polling station where there were like 800, probably only 300 people showed up.
But the machine was broken in the morning.
And it only came back at like 10 o'clock at night.
And they all were waiting there for this entire time in a row.
And nobody kind of cast their votes.
They were all very upset.
Let me tell you this.
Okay.
So, yeah.
There are certain trust assumptions, again, going on here.
Okay.
So, we have the ComElec, the election commission in the Philippines.
They are trusted to scan the ballots in the case vote counting machine.
Which is the machines called VCMs or vote counting machines.
It's malfunctioning.
They are trusted not to misuse the number because they know clearly in which order people registered.
And they also know exactly what the running number is.
So, it's a trust assumption.
And, you know, you have to kind of be clear about it.
Is that something that you as a society want to accept or not?
So, the first one, for example, is culturally.
It could be that in certain countries it actually works.
I'm sure if you would do it in Denmark and they had a machine, which they fortunately don't.
They actually kind of would trust that the election officials would kind of count these things correctly.
As if it is a vote by mail.
Okay.
They are also trusted to protect the paper ballots.
Why is that?
Because they will actually do...
They don't do a risk-limiting audit, but they do something which is called a random manual count.
So, they will actually pick...
That was the thing that Amanda called...
I forgot the second part on her slide.
They pick at random a certain fraction of all of the polling stations and then they recount all of the ballots.
But as you have seen from the ballot going in and out, there are many races.
And there was one race where it's like, pick 11 out of 120 names.
Okay.
So, it's like, how do you recount this?
How do you recount this?
So, they have actually four to five weeks to actually do the recount.
And they will actually open the ballots.
And they will kind of take all of those things, will check that the oval is correctly filled.
There's a rule that it has to be 75% full.
Then it's being recorded, put in a big Excel sheet and counted again.
And then they kind of have a little error percentage and it's very, very close.
Okay.
But you have to protect the paper ballots exactly as in the risk-limiting audits.
If you don't have a secure paper trail, you actually don't have a meaningful audit at the end.
Then the voters are trusted to use the privacy sleeves.
Okay.
But, you know, you've seen that the vendor is trusted to repair broken machines.
And there's the technicians are trusted to scan the ballots four times before determining that it's invalid.
But if you just look at these two trust assumptions, they are somehow contradictory.
Okay.
So, either you have privacy or you have four times and not the technician is doing it.
So, it actually makes little sense.
And so, when you look at this from an election observation point of view,
you can actually kind of say something meaningful out of these contradictions that arise from looking at these trust assumptions.
You could also look at the legal framework.
What the legal framework says is a valid trust assumption and then derive from it.
You know, it's like, okay, is the implementation, is the documents, the tender documents, are they actually consistent with legal law?
And I can kind of tell you about the German case.
This is just a little sidestep here from the Philippines.
But in the German case, they actually changed the law and allowed the use of electronic voting machine.
And there was a person in 2008 whose name I unfortunately forgot who kind of sued in front of the Supreme Court.
And he actually won the case because the election law by itself was actually incompatible with the Constitution.
The Constitution actually requires that every voter should be able to follow all steps of the election by himself or herself carefully.
But it is actually, the law would actually kind of permit the use of technology that actually was much more obscure than that.
And because of that, the election law was invalid.
Okay, these are another thing.
So then there's the other funny thing is about the VVPAT which comes out.
So you have to imagine in the Philippines when they vote, they sit all in a private classroom.
It actually takes a long time to pick 11 out of 120.
And they all come with little papers where they know exactly whom they wanted to vote for.
But then they all sit there like students and fill out those papers for a long, long time.
And then they kind of feed it in and then comes out this VVPAT and they kind of look and compare it very carefully.
Really, they really do compare it and say like, okay, the machine has scanned my choices correctly.
Even if it didn't, it couldn't do anything anyway because they only have one ballot to vote on.
So that's just how it is.
But the interesting thing in terms of vote privacy is they're very concerned about it.
And when they look at their paper, they just kind of look at it, you know, protect it.
And it's like, this is now, this is the real vote.
But the real vote is actually cast in the ballot box.
Okay, so here we also have some, there's something redundant going on because now we have two pieces of evidence.
We have the VVPAT and we have the paper ballots.
Do you need it? No, I would argue you don't.
Okay, but it turns out that in the Philippines, for so many elections, people have actually used those little pieces to kind of convince themselves that they voted correctly.
You can't take it away from them.
But at least you should be aware of it.
The system is unnecessarily complicated, but because of voting culture, it is the way it is and you can't actually get rid of it anymore.
Okay.
Okay, then there is a trusted not to misuse secret number.
That's completely, you know, against verifiability.
All kind of reasonable things of holding an election accountable because can they be trusted not to misuse this?
I mean, it depends on what kind of regime you're kind of running your election, if you can trust them or not.
But if you trust the election officials, then there are other things, you know, you don't actually have to do.
If there's trust already, then, you know, you don't have to make a complicated system.
There's a lot of verification stuff because the trust is already there.
Okay, and the privacy sleeves, I hope I could convince you it's like inconsistent.
Either it's private or it's not private.
I mean, it can't be both.
Okay.
Now I would like to go to Kenya.
Next stop.
Okay.
This is from the 2022 election.
I tried to only pick my pictures where you don't see any faces, so I don't have actually so many without faces.
But here they were all hiding behind the boxes.
Okay, there were like five races at the same time.
When you vote, you get like, they have instead of one paper, they have five papers.
Okay, and they have different colors.
And then you're supposed to be the blue ones in the blue box, the purple, the pink ones in the pink box, yellow in the yellow.
And you can already see that the last two boxes, they are like maybe some color problems that people actually put their thing in the wrong thing.
Then there has to be laws and so on in order to kind of do this.
So, but, you know, I don't want to talk about for Kenya, I don't want to talk about the voting process itself.
I want to talk about the voter identification process.
Because there was really an astounding insight for me, which I had never thought of.
Okay, and that is the following.
So this is a picture which I've stole from the internet from The Standard, which is a Kenyan newspaper.
And so the way how voter authentication actually works in Kenya is they have these devices which are called Kenya integration management kits or system or something.
In short, KEMS. Okay, these are the KEMS devices.
And they have a little fingerprint reader up here.
And so they are clearly used for voter identification.
Okay, and so then they also have, and I didn't want to take any of my pictures because, you know, maybe there's some identifying information about voters here.
So I'd rather give the responsibility to The Standard than keeping it myself.
But here you see a paper backup copy of it.
Now, how many of you believe that it's good to have both?
A paper backup copy and an electronic copy of the voter register.
Who thinks it's good?
Okay, I say it's good.
Okay.
Who thinks it's bad?
You think it's bad?
Yeah, okay.
So, of course, you know, I'm telling you this because it's really, it's really bad.
Okay, it's really bad.
Because, okay, when you see actually here the voters, they're identified by their face with their name and with their passport number or their identification card number.
Okay, so let me tell you how the voting actually really works in Kenya.
So, by the way, this technology was introduced in 2007.
There was an election in Kenya where 500 people were killed and 500,000 people were displaced because, you know, there's a lot of tensions and tribal tensions.
And you can just go back, read the newspaper, but it really was not a good election.
And there was a commission, the Kriegler Commission, who said, like, we need to fix this problem.
And how do we fix the problem in the modern age?
We throw technology at it, okay, in the hope that it does something positive.
And, you know, in 2012 and in 2017, all these issues with the technology.
I mean, all these people died.
In 2017, they killed the IT specialist for the election commission to take his fingerprint somehow, allegedly.
You can read it up in a newspaper.
And then, you know, it was like a big lawsuit.
And in 2017, they actually annulled the election because of that, because the people could not convince others that there was no cyber attacks against the election technology.
Complicated situation.
But anyway, so here we have this.
So this is how it works, okay?
A voter goes in, gets the fingerprint scanned, and the devices actually have the entire database for a particular region of the country, a voting district or something.
So if you're identifying yourself and you're in the wrong polling place, the system will say, you are in room number 17, you should go into room 12.
This is where your voting thing is, okay?
So this is the fingerprint.
Everything is good.
But now, for many people, fingerprint don't work for whatever reason, right?
So in this kind of thing, you actually scan the ID card, okay?
So the ID number seems to be important, okay?
And then you kind of do the fingerprint.
You do that.
It actually selects your record, compares it in the hope that it's better.
If yes, okay, then you get it and everything is done.
And otherwise, it does some supervisor override, whatever.
It's going to be the compare faces and so on.
Everything is good.
In the case that the thing doesn't read, okay, now comes the option where you can manually type in an ID number.
If you only had the ID number of a voter lying around, okay, and then you do the same thing.
And if that also fails, then you can kind of still let the person vote, but you have to kind of fill in the vote.
So if you have the manual register, okay, there's all of the ID numbers.
After polling station closes, what can happen?
It's like there's supposed to be different agents in the polling station that kind of represent the different interests of the different parties.
Ideally.
So if we trust the agents to be honest, here's a trust assumption.
We trust the agents.
Unfortunately, it's the wrong one.
We trust the agents to be honest.
Nothing bad will happen.
If they're, however, in these strongholds, always all collude, okay, then somebody gets this book and then goes, you know, to all of the remaining voters and let them vote.
So there were actually some polling stations in early elections where there was like 99% voter participation in one precinct.
It's like a school.
It's like different rooms next to each other.
One has 99 and the next one has 30% and so on, right?
So something is not 100% there.
So what happens in a country like Kenya?
There was one party said, like, we really need the paper backups because in the case, the system doesn't work.
I mean, it's not just, you know, like, I think both parties did this equally, right?
I'm not accusing anyone.
But one party said, like, we should have the paper backups.
And the other party said, we shouldn't have it because, you know, bad things happen.
Okay.
So then one week before the election, one week before the election.
Yeah.
So first, no, actually, before the election commission decides, no paper backups.
Okay.
Then the one party complains to the courts.
The high court decides one week before the election, we will have paper backups.
Okay.
Then there will be a court of appeal.
The day before the election, the court of appeal says, that was not right.
We will not have paper backups.
Okay.
So it's like, it goes back and forth with the paper backups or not.
So, yeah.
And you could actually easily fix it by just hashing the ID number or doing something else.
Clever business.
Okay.
So, yeah.
So this is the voter identification.
There are now, again, lots of trust assumptions, right?
So the election commission, which is called IAPC, Independent Electoral Boundaries Commission,
is trusted to collect the biometric voter information.
Because in Africa, it's not like here.
You don't have ready databases of all of this stuff.
You actually have to make it for every election.
And they're trusted to keep the BVR up to date.
So half a year before the election, they actually hire a consultancy company to tidy up the election register.
They noticed, I think in 2017, there were more voters on the register than were actually voters in Kenya.
So they had to kind of do something and find duplicates and remove them.
Okay.
Trusted to deliver a paperback of the BVR, including ID numbers.
So that's a trust assumption that they can do it.
And they have always done it, right?
So then the third party entity, trusted to cleanse the BVR, remove duplicates, identification of the deceased, and so on.
Returning officers, who are the people overseeing a polling station, trusted to operate a polling place, of course.
Party agents trusted to represent the party's interest, honestly.
Okay.
Absolutely critical.
And trusted to check signs and result forms.
And, you know, I probably have already given you too much away, but this is a broken assumption.
And when you kind of have this trust assumption, which is broken, actually nothing else makes really sense.
Okay.
Nothing else makes sense.
You cannot... the only argument, and this is what's actually used in order to say, like, the election results in Kenya were approximately correct.
Because they say, like, there are as many strongholds for one as for the other parties, and they all did cheating, so probably it evened out.
Yeah.
So, Estonia.
I don't know how many minutes I have.
Can't be very many.
Okay.
Maybe eight.
Estonia uses Internet voting.
Okay.
So we've heard already this morning from Maya that Estonia is one of the only countries in the world.
It's actually not true.
Switzerland also uses Internet voting.
There are a bunch of countries that use Internet voting.
New South Wales uses Internet voting.
France, I think, has been using Internet voting for return of people living abroad, except they stopped after 2016 for a few years because of hacking concerns and so on.
The interesting thing with the French is always that French always write everything in French, and then it's very difficult to kind of get information, actually, as a non-French speaker about what is going on in France.
I live in Denmark.
It's, like, only, like, around the corner there.
But you really don't know what's going on in France anyway.
So in Estonia, there's an Internet voting system.
And so here, what actually what you have to do is, you know, when you cast your vote, I don't know how much experience you have with Internet voting.
It's a fascinating area.
And also after COVID became actually more important because many are looking at this stuff.
So you're being displayed here a barcode, which you can use in order to kind of verify the vote of yourself.
So it's actually quite easy.
It's actually a hack, OK?
Because what you see there is they use Algamal encryption.
I'm not sure how much you know about this.
But this is the randomness you need in order to kind of make an Algamal encryption.
They just release the randomness of it.
And this is actually enough in order to decrypt the vote.
So when you kind of take an app, you scan this thing.
It goes out to the database, OK?
Retrieves your votes in encrypted form.
And then it uses your randomness that you've used in order to encrypt it, to decrypt it, and shows you the things in clear text.
Yes. OK. So that's very good.
There are lots of things I can tell about this.
But, you know, you just have to read observation reports.
What's really, really important here is actually one thing.
It's like something happened. Something bad happened.
Maybe I should have put the slide first.
Maybe I can just go there.
OK.
OK. So what happened is 63 ballots, coming back to ballot designs, in the election, got the wrong ballot.
How can that be?
It is an Internet voting system, OK?
What actually happens is, in Estonia, it's like when you are a voter, you register in a certain town.
And, you know, and your ballot is for that particular town.
If you move a day before the election, you're actually not entitled anymore to vote on the old ballot.
But you have to kind of vote on the ballot for the next town you've just moved to.
OK. So 63 people got the wrong ballot because something did not go well.
And so what didn't go well is that the updates, you know, the database is being populated from the civil register,
like three weeks before the election.
And then every day, new updates are being installed.
This updating process didn't work.
And they found out about this in the morning of the election.
And so they had all of these updates that they needed to apply.
And unfortunately, this update took longer than 8 o'clock when the election actually opened.
In this, you know, interval of time, 63 voters actually voted and got the wrong ballot.
OK. So now, what do you do in this kind of case?
Now, let me go back to this.
The way how the Internet voting works in Estonia is you kind of vote on your device.
You have a public election key. It's being encrypted and it's being digitally signed.
OK. And then once you have that, you will get all of the ballots.
They're going to be encrypted, decrypted with the private key and the result comes out.
OK. So it's a very simple kind of system.
And here you don't see anything about the verifiability and the zero-knowledge proofs, but they're all there.
And here's actually the thing in detail.
What happens when you are actually collecting and computing the results?
So you have this collector database which has all of these encrypted ballots in it.
So these are pictures that I didn't make.
It's official pictures from the Estonian homepage.
OK. There are timestamps.
You check that the ballots in your ballot box actually are timestamped at the right time.
OK. And then because in Estonia you can vote more than once, with every vote that you cast, you overwrite your previous vote.
And then on the day of the election, you can also vote on paper, overwriting all of your Internet votes.
OK. So you will have to kind of remove votes.
This is why there is a list of voters that the organizers assembled, an annulment list, go into the system and remove them.
Now, the idea of the system is, of course, to only apply this technique to the list of double votes.
The people who overwritten votes or the votes that were overwritten by paper ballots.
OK. But coming back to this little story.
OK. What happened is they decided this is illegal, that these ballots are in the ballot box.
But they know that they have this kind of mechanism.
So they got very lucky that they had this mechanism. Right.
So they asked all of the people to resubmit their ballots.
And 59 actually voted again on the Internet, overwriting their thing.
So the system throws out the old original ballot, which wasn't very valid.
Then there are four ballots that were actually not overwritten.
So there were four ballots that were bad.
So then the committee actually authorized that these four ballots of these four voters are being added by hand into this annulment list.
OK. Just to be thrown out.
And the cleansing process then ran and removed three of them.
And why was that? Is that interesting?
It's because that means that one of these four guys actually must have, or girls, must have voted on paper.
OK. So this is exactly what this picture showed.
That was during the ceremony in the parliament on election night, where all of these votes are decrypted.
And you can kind of see three ballots. Maybe you can't see, but you have to believe me.
I can't read a single word in Estonian, but I think that this is what it means.
It says, like, three ballots were removed in addition to the ones that they actually had to remove.
OK. So here again, there are some trust assumptions.
The civil register is trusted to kind of do a good job.
There's also a question, if something like this goes wrong, who the hell is responsible for this?
Should it be the civil register that should have checked this?
Is it the internet voting vendors that should have checked that the database is actually consistent?
Somebody didn't do it.
But when you be explicit about the trust assumptions, you'll actually figure it out.
OK. Trusted to submit daily updates to the voter register.
The voters are trusted to re-vote if coerced.
That's actually, the re-voting is an anti-coercion kind of mechanism.
You cannot imagine this in the United States, to have a mechanism where you can vote as many times as you want.
Trusted to verify the vote.
OK. This is the barcode that I showed you.
The vendor is trusted to display the correct ballot design.
But, you know, maybe the vendor was actually kind of innocent in this kind of case.
Because the database gave him the wrong thing.
Trusted to cleanse the ballot of overwritten votes.
So, here's the question.
If you can remove any kind of vote, you have to trust the people actually running this technology not to screw up.
That is against verifiable.
Because, how is that verifiable?
So, in the state, the election office is trusted to configure the internet system correctly.
So, anyway, this is unreasonable.
Because people, after many years, only a few percentage points of voters actually check the ballots online.
And trusted is insufficient.
You have to do something else.
OK. So, just as a summary, I'm almost done.
So, trust assumptions is something that we always have implicitly in our head, but we never make explicit.
We need to make them more explicit.
We have to study them ahead of time.
And we have to kind of see what two trust assumptions are complementary, if they're redundant, if you can kind of do it.
Once you know that there's an issue with your trust assumptions, you actually have something to act upon and improve your election system.
And with this, I'd like to just thank you and shut up.
Thank you so much.