By Kim Zetter
Election officials spooked by tampering in a test last week of Diebold optical-scan voting machines should be equally wary of optical-scan equipment produced by other manufacturers, according to a computer scientist who conducted the test.
Election officials in Florida's Leon County, where the test occurred, promptly announced plans to drop Diebold machines in favor of optical-scan machines made by Election Systems & Software, or ES&S. But Hugh Thompson, an adjunct computer science professor at the Florida Institute of Technology who helped devise last week's test, believes other systems could also be vulnerable.
"Looking at these systems doesn't send off signals that ... if we just get rid of Diebold and go to another vendor we'll be safe," Thompson said. "We know the Diebold machines are vulnerable. As for ES&S, we don't know that they're bad but we don't know that they're (good) either."
Thompson and Harri Hursti, a Finnish computer scientist, were able to change votes on the Diebold machine without leaving a trace. Hursti conducted the same test for the California secretary of state's office Tuesday. The office did not return several calls for comment.
Information about the vulnerability comes as states face deadlines to qualify for federal funding to replace punch-card and lever machines with new touch-screen or optical-scan machines. In order to get funding, states must have new machines in place by their first federal election after Jan. 1, 2006.
Optical-scan machines have become the preferred choice of many election officials due to the controversy over touch-screen voting machines, many of which do not produce a paper trail. Optical-scan machines use a paper ballot on which voters mark selections with a pen before officials scan them into a machine. The paper serves as a backup if the machine fails or officials need to recount votes.
The hack Thompson and Hursti performed involves a memory card that's inserted in the Diebold machines to record votes as officials scan ballots. According to Thompson, data on the cards isn't encrypted or secured with passwords. Anyone with programming skills and access to the cards -- such as a county elections technical administrator, a savvy poll worker or a voting company employee -- can alter the data using a laptop and card reader.
To test the machines, Thompson and Hursti conducted a mock election on systems loaded with a rigged memory card. The election consisted of eight ballots asking voters to decide, yes or no, if the Diebold optical-scan machine could be hacked.
Six people voted "no" and two voted "yes." But after scanning the ballots, the total showed one "no" vote and seven "yes" votes.
Diebold did not return several calls for comment.
Thompson said in a real race between candidates someone could pre-load 50 votes for Candidate A and minus 50 votes for Candidate B, for example. Candidate B would need to receive 100 votes before equaling Candidate A's level at the start of the race. The total number of votes on the machine would equal the number of voters, so election officials wouldn't become suspicious.
"It's self-destroying evidence," he said. "Once ... the machine gets past zero and starts counting forward for Candidate B, there's no record that at one point there were negative votes for Candidate B."
Thompson said a second vulnerability in the cards makes it easy to program the voting machine so that it thinks the card is blank at the start of the race. This is important because before voting begins on Election Day, poll workers print a report of vote totals from each machine to show voters that the machines contain no votes.
"The logic to print that zero report is contained on the memory card itself," Thompson said. "So all you do is alter that code ... to always print out a zero report (in the morning)."
David Jefferson, a computer scientist at Lawrence Livermore National Laboratory and chair of California's Voting Systems Technical Assessment and Advisory Board, said that programming software on a removable memory card raises grave concerns.
"The instant anyone with security sensibility hears this, red flags and clanging alarms happen," Jefferson said. "Because this software that is inserted from the memory module is not part of the code base that goes through the qualification process, so it's code that escapes federal scrutiny."
The vote manipulation could conceivably be caught in states where election laws require officials to conduct a 1 percent manual recount to compare digital votes against paper ballots. Parallel monitoring, in which officials pull out random machines for testing on Election Day, might also catch vote manipulation.
But Thompson says machines could be programmed to recognize when they're being tested so as not to change votes during that time. And a manual recount that only examines 1 percent of machines might not be broad enough.
"The question is, if you have altered a memory card in just one of the polling places or even just on one machine, what are the chances that the machine would fall under that 1 percent?" Thompson said. "That's kind of scary."