Securing our votes

This has been a bad few weeks to be a voting machine vendor. Three separate governments, California, Florida and the UK looked at the devices and have come to remarkably similar conclusions. The machines they looked at are poorly designed, poorly implemented and subject to a wide variety of security threats. None of the studies mentioned it, but it is likely that the machines looked great.

The most comprehensive study was done by California Secretary of State Debra Bowen's office. That study looked at three electronic voting systems, each from a different manufacturer. Each system had three separate teams investigating, one looking at the source code, a "red team" that had physical access to the device and an accessibility team. Their conclusions were not surprising to anyone who has paid attention to this issue over the years.

All three of the voting machine systems were found to be sorely deficient by all three teams. Even accessibility, which is one of the major benefits touted by electronic voting advocates, was found lacking:

Though it is certainly terrible not to meet the needs of some individual voters, safeguarding the election process and accurately reporting the vote totals need to be higher priorities. Since they obviously had not successfully completed the accessibility task, one would hope they were able to secure the voting process. Unfortunately, they could not get the primary job done right either.

The red team reports were released first and the conclusions were devastating:

The teams were able to defeat the physical security of the voting machines, modify or overwrite the software in the machines as well as subvert the tabulation machines in order to provide incorrect vote counts. All of this just by having access to the machines themselves; the same access that election officials, poll workers and, to a lesser extent, voters, have.

Several days later, the source code teams' reports were released and, at that point, were almost anti-climactic. Unsurprisingly, they found numerous, hideous source code flaws in all three systems. Buffer overflows, hard coded passwords ('diebold' being a particularly difficult one to guess), misuse of encryption, integer overflows (wrapping vote counts to negative or zero perhaps); the list goes on an on. It is as if the voting machine vendors are completely unaware of the last twenty (or thirty or forty) years of software security flaws.

In reality, they are most likely not unaware, they are just arrogant. Diebold, Hart and Sequoia (the companies whose machines were studied) do not depend solely on their technical "prowess" to win bids for providing voting machines, politics plays a huge role. These are well connected companies. It also helps that they are all uniformly bad, there are literally no secure choices for a government agency to make.

Florida's study only covered Diebold equipment, but it echoed the findings in the California study. Avi Rubin of Johns Hopkins University, who participated in a 2003 study of Diebold's voting machine, notes:

One of the bigger problems found was that Diebold assigned cryptographic keys to each voting machine that is derived from an MD5 hash of the machine's serial number. Rubin again:

The UK also released reports on the outcome of electronic voting trials held in May. The overall summary of the trial, was, once again, not very favorable:

This was not the result of security professionals analyzing the systems for flaws, but was instead noted in actual trials of the equipment in an election.

The California study was quite well done and well thought out, except for one thing: it was done long after the equipment was bought and used in elections. This is the kind of study that needs to be done before buying the equipment. Due to the conclusions of the study, Bowen revoked the certification of the equipment from all three vendors, but immediately had to conditionally re-certify them as a practical matter. Even with a six month lead time, replacement systems (either electronic or of some other kind) could not be deployed before the 2008 California presidential primary voting.

The reaction to the California study by the manufacturers was typical. It is the same reaction they have had to each and every study done of the security of their devices: trivialize it. Each released a statement in reaction to the study conclusions, essentially admitting the flaws, but claiming that any "laboratory study" would find vulnerabilities. According to these vendors, it is impossible to make a secure voting system.

As they certainly know, no one is asking these vendors to break the laws of physics or to produce perfectly secure code. It would appear that they expend far more effort in deflecting criticism and lobbying various legislative bodies than they spend trying to secure their code and equipment. It is not necessary that the equipment be tamper-proof, merely that tampering can be detected. At least minimal precautions, perhaps to the level taught to computer science undergraduates, should be taken with the software.

This is not anywhere near as hard a problem as the vendors make it out to be. Many of the techniques needed to secure voting machinery are well known and well understood, at least outside of the vendors' labs. This is an area where open source methods could be and should be applied. Organizations like BlackBoxVoting.org and the NSF Accurate project should be working on solutions. Private companies have shown themselves to be completely incompetent at producing secure voting equipment, it is time for another solution to be tried.

Note: you can avoid this step in the future by logging into your LWN account.