only hope that future designers of socialbots with different goals will adhere to Asimov’s Three Laws of Robotics [ 7] to reduce the risk of harm to the human systems in which they operate.

@tinypirate is a government drone in New Zealand who spends his spare time concocting schemes and convincing smarter friends to act on them. @AeroFade is a computer security researcher from New Zealand with an Hons. in computer science, interested in studying how online culture shapes the offline world.

The Socialbot Network:
Are Social Botnets
Possible?
Yazan Boshmaf, Ildar Muslukhov,
Konstantin Beznosov, and
Matei Ripeanu

March + April 2012

interactions

Online social networking services (SNSs) have far exceeded their original goal of connecting friends, family, and acquaintances. Today third parties use SNSs as an effective medium to reach out to millions of active users via social media campaigns. The effectiveness of such campaigns and the long-term survival of SNSs rely on the trust among these users, which is materialized through publicly exposed social connections (e.g., friendships on Facebook, follower/followee relationships on Twitter).

A new attack vector on such networks thus becomes possible: A malicious entity that not only controls a large number of SNS profiles but also establishes an arbitrarily large number of connections with human users can threaten the long-term health of the SNS ecosystem.

To counter this threat, today’s SNS security defenses block hijacked SNS accounts that are usually controlled by spam bots. Such defenses flag accounts as malicious based on their behavioral patterns. However, the robustness of these defenses against socialbots—automated profiles designed to mimic human behavior—is relatively unexplored.

To fill this gap, we adapted the design of existing botnets and built a socialbot network (SbN), a group of programmable socialbots that are controlled by an attacker using a software controller called the botmaster [ 8]. We deployed our SbN prototype, consisting of 102 socialbots and a single botmaster, on Facebook for eight weeks during the spring of 2011. We selected Facebook as the target SNS for two reasons: It is the largest SNS today, and it represents a friendship network where users connect mostly with friends and family but not with strangers. Overall, the socialbots sent 8,570 connection requests, out of which 3,055 were accepted.

Our experiments yielded mul-
tiple findings. First, we demon-
strated that SNSs are vulnerable
to large-scale infiltration. Not
only is it feasible to automate the
operation of an SbN with minimal
resources, but users’ behavior
in SNSs can also be exploited to
increase the likelihood of a suc-
cessful infiltration. For example,
we observed that the more friends
a user has, the less selective she
will be when screening out friend-
ship requests sent by a socialbot.
Moreover, users are even less
selective when they have mutual
friends with socialbots, when the
chance of accepting a friendship
request from a bot reaches up to
80 percent. Second, and equally
important, bots that mimic real
users (e.g., by posting intriguing
status updates crawled from the
Web) make it difficult for other
users and SNS security defenses to
identify them as bots.