The purpose of an election is to:
- fairly decide a winner; and
- convince everyone that the decision was fair.
If anyone can exploit a security vulnerability in voting machines or online voting systems, they can modify the individual votes or the final count such that the wrong candidate wins. There are some insurmountable theoretical security issues with electronic voting which can never be resolved. Worse still is that in practice voting machines and online voting are consistently insecure against even the most basic attacks, due to a shocking lack of due diligence by many different electoral organisations. This means that electronic voting is far less likely to correctly select the winner than paper counted manually.
Not only is election fraud easier to execute with electronic elections, but it is also harder to detect. The stakes for national elections are very high, and the threats are very real. Elections which lack a paper trail cannot be re-counted. Voters are typically not allowed to inspect and verify the software and devices used to log and count their votes. Instead they must blindly trust all the people in the supply chain, and each electoral official. Even if inspection was allowed, it is scientifically impossible to confirm that any non-trivial electronic device has not been infected with malware. This means that the integrity of any electronic election is not trustworthy. There is not much point in holding an election unless the people trust the outcome, and believe in the process.
Comparing Banking and Voting
Many people ask the question:
"If I can securely do my banking online, why can't I cast my vote online?"
The short answer is that:
- banking and elections have two very different threat models; and
- elections have a unique anonymity requirement.
Cryptography is not the primary defence which keeps your money safe when banking online. The primary security measure is the fact that when your account gets cracked, your bank will just pay you back, out of their own pocket. They will cop that as an operational loss, because doing so is cheaper than implementing extreme security measures.
One big difference in requirements between commerce and voting is that of anonymity. An electoral system must make it impossible for anyone to find out how you voted, even if you want to show them. If demonstrating how you voted were possible, voters could be coerced. (e.g. "Vote for Party X or else I will hurt you.") If it is impossible for you to show them how you voted, then you can still vote how you want and just lie about it afterwards, thus making coercion ineffective.
An electoral system must separate your identity and the fact that you have cast your vote, from the content of your vote. Banks never have this problem. They never want to authorise a transaction without simultaneously knowing how much money is moving and who the parties are. Consequently bank fraud is very easy to detect. Either:
- the total account balance does not equal the sum of all past transaction; or
- there is a transaction which the account owner does not recognise.
There are several types of electronic voting:
- Fully online voting (using a website or app on your own device)
- Paperless voting machines in poll booths
- Machines which count paper ballots (and possibly a second machine to print the paper)
For the online option you could implement a verification system, so that after casting your vote from one device, you log in on another and ask the server to tell you what it thinks your vote is. If this is possible then your vote is not truly anonymous, so coercion becomes possible, which is not acceptable.
For the first two options (which are paperless), it is literally impossible to detect most fraud if it happens, unless you give up anonymity. Of course election fraud is sometimes easy to detect, like the time multiple vote counting machines submitted negative votes for Al Gore. However it only takes a dozen flipped votes at a carefully chosen electorate to change the outcome of the whole election. How could you detect that? If the outcome is a close call, or the integrity of the machines is in doubt, you cannot do a re-count. The possibility of a recount is one of the main essential features you lose if you abandon paper.
Some "VVPAT" systems use a digital touch screen for voter input, and then print a paper ballot, which is mechanically dropped into a sealed box. This allows voters to see their vote printed (without touching it), to ensure the machine did not change their vote. However what happens if a voter claims the vote they saw printed was not what they selected? Do you stop all voting on that machine, all machines in that polling place, all machines in the country? What if the voter just mis-remembered what they chose? What if the voter lied in a deliberate attempt to disrupt voting in a crucial electorate? There is no way to know.
Some systems use a voting machine to provide a simpler and nicer interface than paper. They print a ballot, which the voter can inspect and then insert into the box by hand. However in practice half of voters don't check, and most who do check don't notice errors anyway. Ultimately such a system is just a very expensive pencil.
It might seem like this does not matter if you just make sure that the devices don't get compromised in the first place. However recounting is a safety net which provides an extra layer of security. All good security systems use multiple layers of protection for Defence in Depth (like multiple concentric walls protecting a castle). Furthermore as enumerated below, it is impossible in practice and even in theory to defend against the kind of bad actors which elections face.
A bank only needs to defend themselves against script kiddies, and at worst organised crime gangs. Election officials must defend their system against foreign nation states. Their enemies are as powerful as the NSA, and Russia's Kremlin. The stakes for big elections are literally trillions of dollars, so the enemies will be willing to spend comparable amounts of money to change the outcome. Examples of attacks which are only used when nation states are involved and the stakes are this high include:
- the STUXNET attack, where the US and Israeli governments managed to inject malware into a computer which was completely offline, by using infected USB drives, and multiple zero-day exploits
- the time America's NSA deliberately lobbied standards group to adopt an encryption algorithm they already knew how to break, which was subsequently used in important systems all over the world
- the Stingray attack, where governments mimic mobile towers to execute Man-In-The-Middle attacks on the 2G network. (Many voting machines transmit vote totals over the 2G network.)
- Governments can use laws like Australia's Assistance and Access bill to force technology companies to secretly install backdoors in their systems, for products which end up being used as part of a foreign elections. (Government's around the world are banning Huawei routers and devices from critical infrastructure, solely because they are worried China will use this exact attack vector.)
Scariest of all is the Ken Thompson Hack. Someone may write innocent and strong software for a voting machine, but if that software developer's laptop was infected with this kind of worm, the compiled (computer readable) version of their code can be modified to be different to the human readable code which they wrote. For example, the worm could tell the voting machine to change every tenth vote for Party X to a vote for Party Y. The shocking thing about this is that since it is impossible to convert any non-trivial computer code back to human-readable form, this attack is undetectable. Since some software on every computer originated from another computer (which could be infected), it is theoretically impossible to know with certainty that any device has not been infected. This attack is difficult to execute in practice. However when trillions of dollars are on the line, and large foreign governments want to sway the outcome, it becomes a realistic threat.
Banks do not need to worry about these kinds of attacks, because their attackers are less formidable.
Electronic Voting In Practice
The above section explained how the Ken Thompson Hack means that there exists malware which is impossible to detect or prevent. However in practice, the actual state of online voting and voting machine security is far, far worse. Electronic voting systems are consistently designed and operated very insecurely again and again and again.
"What I’ve seen in the past ten years is that the vendors have absolutely fumbled every single attempt in security" - Jacob D. Stauffer, who has conducted voting-machine security assessments for California's secretary of state for a decade
At DEF CON (an international security convention) voting machines were made available to attendees to see how well they stand up to actual attackers.
Thirty minors were able to compromise the machines, including an eleven year old who changed the outcome of the mock election in ten minutes. Machines which can be compromised by an eleven year old in ten minutes are being used for real elections with actual stakes.
There are many examples of real elections where voting machines were wirelessly connected to the internet, using weak protocols, running remote desktop software (which allows remote users to control the device).
If you can think of a way voting machines or online voting could go wrong, then there is almost certainly an example of an actual real election where exactly that has happened. Some real elections have used machines which:
- transmit vote counts to the central office without any encryption (This is like stapling a $100 note to a postcard and expecting it to get to the destination.)
- download software updates without any encryption, and without any cryptographic signatures, from servers which can be easily taken over
- run unpatched remote desktop software, with known vulnerabilities
- run versions of Windows which were abandoned by Microsoft over a decade ago, and have received no security patches since then
- use weak passwords like "admin" or "test", or hard coded passwords which can't be changed, or the instruction manuals recommend using and re-using weak passwords
- use passwords which are saved in a file which is publicly accessible online
- use Wi-Fi, with weak, outdated encryption algorithms which can be cracked from outside the building in ten minutes
- contain undocumented SD card ports and exposed USB ports which voters can plug malicious drives into
- save no logs, so the modification of data or simple changing of software would not be detected
- are sold without having onboard voter data erased, and with the "Property Of" government labels still attached (akin to selling a police car with the police logo still on it)
- use easily-pickable physical padlocks, and contain chips which can be reprogrammed by just plucking them out and replacing them with chips with other code
There was even a case a stray radiation particle from space hitting a transistor and flipping a single bit, which changed the numbers, causing the other candidate to win. This was only detected because the new total count was too high, by an amount which was a power of two.
That is what happens with devices designed and used solely for elections, secured by dedicated electoral organisations whose job is to keep the devices secure. Online voting where voters can cast their vote using an app or website on their own consumer device will necessarily be worse, because consumer devices are less diligently secured and contain a lot more code and features. Many actual elections which have used online voting have had real production servers taken over by attackers, easily.
"Within 36 hours of the system going live, our team had found and exploited a vulnerability that gave us almost total control of the server software, including the ability to change votes and reveal voters' secret ballots." - Prof. J. Alex Halderman, computer scientist
In addition to viewing and modifying votes, these researchers left a deliberately obvious calling card, by making the voting page play their Universities fight song. The officials didn't even notice this deliberately obvious attack until two days later.
There is at least one example of a real election server being attacked by a malicious attacker using SQL injection. This is the oldest trick in the book. Such vulnerabilities are so well known and easy to prevent that it would be shocking to see on a read-only blog, let alone an electoral system.
Australian elections mostly use paper, but the state of New South Wales offers an online iVote system for overseas voters. The makers of this site chose to load a third party analytics tool inside the vote page. Such tools are designed to provide insight about visitor behaviour to website operators. (e.g. "40% of visitors leave before scrolling to the bottom of the page.") These tools are nominally passive, however it would be trivial for that vendor to modify their code to read your vote, modify anything on the page, or even modify your vote directly. This code was loaded insecurely, so anyone who compromises this monitoring tool could change visitors' votes. Furthermore, the main iVote home page prior to the actual ballot page was not encrypted at all, so an attacker could just hijack that page and then direct you to a malicious page which looks like the real one and still makes your browser show a green padlock icon.
Most consumer devices (such as phones and laptops) are already infected with relatively simple malware. If consumers cannot protect themselves from malware made by simple individual attackers, how could they possibly defend against malware made by a foreign nation's military, specifically for the purpose of election rigging? If you cast a vote on a device which is infected with something, you cannot know that what you see on the screen matches the data which is actually sent to the server.
Computers can perform checksums and hashes to "fingerprint" their software and ensure it matches the official fingerprint. However if the device is infected such that an attacker can change your vote, then it is easy for them to also change the fingerprint calculation, thereby hiding their tracks.
The only way to mitigate this is to allow users to subsequently request a confirmation of who they voted for (such as from a different device). If this is possible then your vote is not anonymous, so coercion becomes possible.
Furthermore, it is technologically impossible to meaningfully separate the fact that you voted (to prevent double-voting) from who you voted for (anonymity).
How can election officials be sure that people who cast their vote on their laptop at home did not have an abusive partner looking over their shoulder and forcing them to vote for a particular candidate? When voting only happens in private booths inside public rooms, this is not a problem.
How Big a Flaw is Big Enough?
You only need to fudge a few votes in a carefully chosen electorate to sway the whole election. You need to modify even fewer to destroy trust in the system.
Could a Blockchain Fix Security Issues?
Blockchains do not make elections more secure. Many people such as The Flux Party have drunk the blockchain Kool-Aid and now throw around the buzzword without any pronoun as if it will magically fix all problems.
All software contains bugs. Blockchain software is consistently worse than normal software. Blockchain apps are typically written in Solidity, which is one of the most awful and error-prone programming languages of all time.
The very reason blockchains were invented is antithetical to elections. Blockchains were designed to be public, immutable and fully distributed. Normally with blockchain applications, if someone steals your password they permanently take control of your account. For an election this means someone else can spend your vote for this election and all subsequent elections. If it is possible for the electoral agency to perform a password reset, then a single centralised party has the ability to block and modify writes to the database. Blockchains were invented to prevent exactly that from being possible. So if password resets are possible, there's no point using a blockchain.
Could Regulation Fix Security Issues?
As explained above, the Ken Thompson hack means that some malware can infiltrate systems and remain undetected by even well funded, well trained people using world's best practice. All the practical implementation flaws enumerated above also remain despite regulation.
In most US jurisdictions it is illegal to connect machines to the Internet. However vendors consistently use channels which are equally or less secure, such as unencrypted 2G phone calls, vulnerable to man-in-the-middle attacks.
It is typically illegal to install remote desktop software on voting machines in the US. However one vendor did it anyway, and then lied about it. The software they installed allows remote execution of arbitrary code. This tool was so insecure that the company who makes it recommends not using it. The machine makers claim they had no knowledge of that software, so how did it get there? How can you patch code you don't know is there? Either the machine makers were malicious or incompetent. Both possibilities are bad.
Many of the prior examples were from the US because they are one of the biggest users of digital voting systems. However these issues have cropped up in other regulatory jurisdictions too, such as Germany, Estonia and Australia.
In practice voting machines are typically regulated, audited, inspected and secured less diligently than gambling machines.
Commercial Vendors and Proprietary Software
It only takes a few lines of code to turn innocent vote counting software into malicious vote-flipping software. The first step for any regulator would be to insist upon the use of only public, open-source code. That way regulators and voters could see that the code which is supposed to eventually end up counting votes is not malicious. However this never happens. In practice voting software is always proprietary. This means it is secret. Vendors refuse to show the code to security researchers, let alone average citizens. In some cases it is even illegal for security researchers to try to probe such software for vulnerabilities. (Such laws won't stop bad guys from finding vulnerabilities to rig the election with, but they will stop good guys from finding those vulnerabilities and fixing them first.)
Recall the second reason why elections take place. Voters who support the losing party must be convinced that the decision was fair and democratic. If you have voting machines, this means that they must be convinced the machines were not compromised. How can voters trust the election results when the method of counting and verifying their votes is secret?
Imagine if you cast your vote by making a phone call to a human, telling them your vote. Then at the end of the day that person calls the central office and says what their totals were. That's crazy. That's insane. That's obviously a terrible, fundamentally fucked idea. But that's effectively how all proprietary voting software works.
In the 2016 Australian federal election ballots were counted by machines inside the vendor's building. The vendor refused to let most official scrutineers in to see all the counting machines operating, instead only allowing a small number of scrutineers watch a minority of the machines. The scanning performed by Fuji Xerox machines was 'checked' by Fuji Xerox employees, in only a few seconds per ballot. For the few ballot scans which were seen by scrutineers, if scrutineers objected after the vendor staff already clicked next, it was impossible for the staff to go back to that prior ballot and correct it. Thankfully the electoral commission ended up recounting all ballots by hand. However they never bothered to figure out just how many mistakes were made by the machines.
By choosing to use proprietary voting systems, governments are giving control of the election to a private company who have a strong financial incentive to hide known vulnerabilities and prevent security researchers from finding bugs.
All software has bugs (coding mistakes), typically around ten to fifty bugs for every thousand lines of code. Voting machines typically have several hundred thousand lines of code. Therefore we can say with high confidence that any voting machines probably has thousands of bugs. The consequences of using software with so many bugs to run an election are catastrophic. The video above is from a real election. This example is one of many where the screen displayed a different vote to what the voter intended. For online voting, votes are cast on consumer devices running tens of millions of lines of code, which means they contain hundreds of thousands of bugs. These problems do not happen with paper ballots.
In March a dozen European Union members cast their vote incorrectly, because the user interface of the voting systems did not make it clear whether they were voting on amendments or the whole bill. The consequence is that legislation not supported by the majority was passed, and Internet freedom all around the world was impacted as a result. This happened to politicians whose entire job is voting, and understanding the process. So how could an average person voting once a year or less possibly fare any better?
In one Texas election the machines took many seconds to render each page on the screen. If voters touched the screen before that rendering finished, the machine silently threw away their vote. The officials' response was to claim nothing was wrong, and to blame voters for not waiting "at least three to five seconds for all choices to be rendered". If a computer takes five seconds to print some simple text on a screen from a local drive, then something is fundamentally wrong with how that software was written.
Paper ballots are incredibly cheap. The ballot costs a few cents, and the cardboard booths and pencils cost a few dollars. So polling places can afford to install as many as they can fit into the room. In Australia many voters can walk straight in, and it's rare to queue for more than twenty minutes. In contrast, voting machines cost several thousand dollars each, so there tend to be fewer booths per polling place. Consequently voter throughput is lower, so voters frequently queue for many hours. This discourages voters, which is bad for democracy.
The Benefits of Paper
In the 2000 US presidential election paper ballot counting machines miscounted a substantial amount of votes, because of the hanging chad issue. The consequence was that the elected president was quite possibly not the president who received the most votes. This would not have happened if the ballots had been marked and counted by hand.
Rather than go back to manual counting the US doubled down on electronic voting, of all types. It is now common for elections in the US to use two stage systems with paper in between, or direct vote systems with no paper, or even email ballots. (Most email is completely unencrypted. It is trivial to send an email and make it look like someone else sent it. So this is as secure as voting via postcard.)
The method of physical voting with paper ballots and manual counting is centuries old. Pretty much all attacks have already been tried, discovered, and defended against. For example, the reason you are given a pencil not a pen is to defend against the possibility of voters swapping pens for ones with disappearing ink. That is the level of paranoia which electoral systems should have.
Even if someone does manage to commit election fraud with paper, the impact would probably be constrained to just one ballot box or just one polling booth. Whereas if an electronic election is compromised, the attacker can influence the entire count. For example, many voting machines have the candidate list loaded on each election using an SD card, which is inserted into all machines, one after the other. It is relatively simple to construct a virus such that if you infect just one machine, any inserted SD card will be infected, which will infect all subsequent machines programmed with that card.
Paper ballots are typically counted by you, ordinary citizens. The staff and scrutineers are people from across the political spectrum. In contrast most voting machines are made by only a handful of vendors, who are highly partisan. The CEO of one publicly proclaimed that he will be "helping Ohio deliver its electoral votes to the President."
If you are concerned that the people counting your vote will deliberately fudge the numbers, you can simply walk into the polling place and watch them, or participate in the count (if you register in advance). If you are concerned about a voting machine, there is nothing you can do.
For systems which have no paper trail, there is no possibility of a recount. You are forced to blindly trust the integrity and competency of everyone in the supply chain, which includes:
- upstream hardware makers overseas
- everyone who has access to the machine in the years between elections
- every voter who had private physical access to the machine when they voted
and so on.
For systems which print out a paper receipt for you but submit the vote digitally, you cannot know whether the machine flipped your vote in the digital record, but printed your intended vote on the receipt.
For ballot counting machines, what happens if you want to recount because you question the machine's integrity? You cannot recount with the same machine, because that will give you the same possibly-corrupt answer. What about another machine in the same polling place? If the first machine was infected, that one could be too. What about another machine in another polling place? That's probably made by the same vendor, with the same flaws.
It only takes a few extra bytes of code to turn legitimate vote counting software into vote-flipping software. It is impossible to convert non-trivial ones and zero back into human readable form. Voting machines are a black box.
One of the challenging problems which paper solves is ensuring that each person votes only once, that vote definitely ends up in the count, and no one can see how any individual voted. This works well with paper ballots, marked by hand. No other system has been invented which satisfies this constraint well.
One significant motivation for electronic voting is to make the counting process faster. But why do we need to rush? In the 2016 Australian federal election it took 4 weeks until results were finalised. The nation did not falter. The sky did not fall. Parliament steers the nation on a time-scale of years, so waiting hours or even weeks for a count is not of great consequence.
As mentioned earlier, voting machines are expensive. In the 2016 Australian federal election, distrust in the vote scanning machines resulted in a manual recount, at a cost of $8.6 million. That is less than the cost of merely transporting the ballots to the digital scanning centres ($8.7 million). The cost of designing and building the digital scanning system was $27 million. So cost reduction is an invalid motivation for electronic voting.
Another motivation for online voting is convenience for voters. As mentioned earlier, voting with paper is quicker than voting with machines. So the only convenience increase for voters is if they are out of town on election day and they can vote online. Such voters can already cast a vote in the weeks prior. This is increasingly popular, with three million voters doing so in the 2019 election. Convenience is important. However online voting violates both of the original goals of an election (fair decision, which everyone trusts is fair), and we already have workarounds for people out of town on election day. Citizens in Australia cast a vote for any level of government about once per year, and the stakes are very high. Therefore the trade-off is not worth it.
The purpose of an election is to fairly choose a winner, and convince everyone that it was a fair decision.
Electronic voting has fundamental theoretical security issues, so you can never be sure the decision was fair. In practice voting machines and online voting are consistently implemented with shocking security, far from best practice. Every vulnerability you can think of has probably been found on a real system.
Paper voting and manual counting has been used and refined over centuries. Electronic voting is more expensive, and involves handing control of the election to a partisan for-profit vendor who has a strong incentive to hide security vulnerabilities and reduce cost by cutting corners.
There is no real need to expedite the process. The sky doesn't fall if counting takes a while.
If it ain't broke, don't fix it. Electronic voting isn't even a fix. Time and time again it has failed to even come close to the security and usability of paper ballots, counted manually.