Old lever voting machines had a simple mechanism for preventing voters from casting multiple votes. Once a levers was flipped, it wouldn't reset until the voter pulled the privacy lever to leave the booth. One person, one vote. Vote audits function as a sort of post-election security: They can be used for recounts or to verify votes, but can't prevent fraud while votes are being cast. So how do modern electronic voting systems prevent tampering?
Voting machines have been scrutinized and researched for security flaws and potential issues. But not every potential issue produces change: For example, as mentioned on the DRE voting machine page, networked machines use modem connections even though hacking that data is theoretically possible. Data storage is typically quite secure. Diebold voting machines, for example, store votes on two separate memory units and the machines have power backups to prevent data from being lost. Still, there are cases of vanishing votes. An Election Systems and Software, Inc. machine used in a 2002 Florida election crashed -- and deleted election results.
When Australia was experimenting with electronic voting machines, their solution to security concerns was simple: Let everyone scrutinize the software. They made the code open source so that anyone could hunt down flaws [source: Wired]. Unfortunately, many studies of electronic voting machines have revealed both physical and software vulnerabilities. The 2006 report The Machinery of Democracy examines DRE and optical voting machines used in the U.S. Long story short: The right skills and the right access definitely make it possible to tamper with machines. So what happens when things go wrong?