A small array of scripts programmed to pass themselves off as real people stole 250 gigabytes worth of personal information from Facebook users in just eight weeks, researchers said in an academic report to be presented next month.
The 102 “socialbots” included a name and picture of a fictitious Facebook user and used programming interfaces from ihearthquotes.com to automatically embed pseudo-random quotes into status updates. They also used Facebook interfaces to send connection requests to about 5,000 randomly selected profiles. They then sent connection requests to the friends of those who accepted the initial invitation, and with each acceptance, they scraped whatever information was available.
At the end of the eight-week experiment, the researchers made off with 250 gigabytes of personal data, much of it configured to be available only to people on the user’s list of friends.
A defense known as the Facebook Immune System, which is designed to automatically flag fake profiles, did little to thin the army of socialbots used in the study. While about 20 percent of them were blocked, the closures were the result of feedback from other users who reported spam, the researchers said. Their socialbot network targeted Facebook, but they said similar ones could penetrate virtually any OSN, or online social network.
“As socialbots infiltrate a targeted OSN, they can further harvest private users’ data such as email addresses, phone numbers, and other personal data that have monetary value,” the researchers, from the University of British Columbia Vancouver, wrote in the paper (PDF), which is scheduled to be presented at next month’s Annual Computer Security Applications Conference in Orlando, Florida. “To an adversary, such data are valuable and can be used for online profiling and large-scale email spam and phishing campaigns.”
During the initial “bootstrapping” phase of the experiment, the socialbots sent friendship requests to 5,053 randomly selected Facebook users. To prevent the triggering of fraud detection systems, each fake account sent only 25 requests per day, a constraint that required two days for all of them to be processed. Within two weeks, 976, or about 19 percent of the requests, were accepted.
Over the remaining six weeks, the bots sent requests to the Facebook friends of those who accepted the initial invitations. Of the 3,517 users who received the second round of requests, 2,079, or about 59 percent, accepted. With further refinements, the socialbots could achieve a large-scale infiltration with a success rate of about 80 percent, the researchers said.
The significant jump exhibits what researchers call the “triadic closure principle,” which predicts that the likelihood of someone accepting a connection request in a social network is about three times higher when the pair has mutual connections. This principle proved to be a boon to the socialbots in another respect: they received 331 requests from Facebook users in the socialbots’ extended neighborhoods.
A Facebook spokesman declined to comment on the report.
“However, we always remind our users to only accept friend requests from those they know and trust,” he wrote in an email to The Register. “We use a combination of three systems here to combat attacks like this – friend request and fake account classifiers, and rate-limiting techniques. These classifiers block and disable inauthentic friend requests and fake accounts while rate-limiting truncates the damage that can be done by any one entity.”
Besides stealing gigabytes worth of pictures, phone numbers, and other data, socialbots could be used to generate comments that are designed to appear as if they spontaneously came from thousands of individuals, when in fact they are an astroturf campaign that’s the work of a single actor. The computer worm known as Koobface already uses compromised Facebook accounts to trick friends into installing malware on their computers. Other socialbots are sold online for about $29 apiece, the researchers said.
The researchers behind the army of socialbots include Yazan Boshmaf, Ildar Muslukhov, Konstantin Beznosov, and Matei Ripenu. In an email, Boshmaf said their objective was to improve the security of privacy of social networks.
He said: “Overall, our research goal is not to expose Facebook Immune System’s vulnerabilities per se, but to help Facebook and the wider community to build more secure systems that are less vulnerable to both human exploits (i.e., social engineering) and technical exploits (i.e., platform hacks).”