A study of the ZeroAccess botnet and its ability to use fraudulent advertising clicks to generate revenue demonstrates that online ad networks have insufficient countermeasures to combat the cyber-criminals who abuse them, a group of nine academic researchers concluded.
The research focused on ZeroAccess as perhaps the best known click-fraud botnet. Click fraud uses compromised systems to click on advertisements, earning advertising affiliate fees for the criminals operating the botnet. In their study, the researchers were able to identify 54 ad units, or campaigns, related to ZeroAccess that produced about 1 million fraudulent clicks per day, with a likely value of $100,000.
Because advertising networks use a wide variety of intermediaries—such as syndicators, traffic resellers and aggregators—tracking fraud using the network is very difficult, Vern Paxson, professor of electrical engineering and computer science at the University of California at Berkeley, told eWEEK.
“Unfortunately, one of the findings from our work is that the ecosystem is hugely tangled, and sort of beautifully designed in a way to launder identity and provenance,” he said. “And therefore it is really hard to map it or come to a sound estimate of financial scale.”
In the paper presented at the 21st ACM Conference on Computer and Communications Security in Scottsdale, Ariz., Nov. 3-7, Paxson and other researchers from the University of California at Berkeley, Microsoft, George Mason University and the University of California at San Diego released details of their study, conducted with data from a cooperating advertising network.
The researchers attempted to use a number of techniques to tease out information on the level of click fraud occurring on the advertising network by analyzing the click data from the researchers’ ad partner, as well as ZeroAccess DNS data and other telemetry.
While a major event—the December 2013 takedown of ZeroAccess by Microsoft and its industry and law-enforcement partners—led to a drop in click fraud, analyzing the click-fraud data and the event data did not determine a fail-proof way to accurately classify good and bad advertising clicks. The study, however, did identify 54 ad units that an aggregation of evidence suggested were fraudulent.
The business relationships between publishers, advertisers and every business in between are so convoluted that ZeroAccess clicks are blended with legitimate clicks and clicks from other fraudulent campaigns, creating a fog of war that makes attributing the source of clicks seemingly impossible, Paul Pearce, a co-author of the paper and a Ph.D. student at UC Berkeley, told eWEEK.
“The blending can be a combination of people who are intentionally trying to obfuscate the source of traffic, but also may just be traffic brokers who deal with so many different sources that, by the virtue of how the system works, it gets blended in with legitimate things,” Pearce said.
The study suggests some ways to improve defenses against click fraud.
One way to discourage click fraud is to make sure that publishers that engage in such fraud lose legitimacy. Unknown publishers have no reputation to lose and so the penalties of click fraud typically fall flat for most sites. This is particularly true for publishers that join a syndicate, which provide advertisements and then return a portion of the profits back to the publishers. Such publishers are effectively anonymized by such services, according to the researchers.
Yet, a technique known as smart-pricing has been somewhat effective in reducing the cost of fraudulent clicks, the researchers found. Smart-pricing reduces the price paid to a publisher if ad clicks from the site are less likely to result in a conversion, such as when a consumer takes some action desirable to the advertiser. The researchers found the smart-pricing does not eliminate click fraud, but mitigates the cost of fraudulent clicks.
“The complexity of the online advertising ecosystem is such that no one party comes remotely close to having comprehensive visibility into who gets paid what for any given click,” the researchers conclude in the paper.
“The hodgepodge of data sources required for our analysis starkly illustrates both the tangled nature of the click fraud problem space and the pressing need for much better mechanisms for correlating traffic and payment streams.”