Securing Facebook: With a Little Help From His 800 Million Friends
The eyes of the online world are on Joe Sullivan.
As the CSO of Facebook, Sullivan is without a doubt one of the most visible security chiefs in the business. He must mitigate myriad security and privacy risks not only for Facebook's employees and corporate systems, but also for the social network's 800 million members.
Sullivan, 44, joined Facebook in 2008. He moved to the private sector 10 years ago to focus on security, and before that he was a federal prosecutor for eight years. His legal background has come in handy of late, as Facebook has sued several people for misusing the service.
CSO contributor Lauren Gibbons Paul talked to Sullivan about the challenges of managing security on a rapidly evolving social network.
CSO: With all the publicity about privacy and security regarding Facebook, what do you regard as the biggest threats?
Joe Sullivan: I think the challenge with being at Facebook is that it's always about trust. People need to feel secure when they use Facebook. When it first came along, people were not comfortable putting their photo and real name on the Internet. But that's the way Facebook works-it's your real name and real identity interacting with real people in your life. If [members] experience something that erodes their trust in that experience, they're not going to come back.
[So] we have to invest really heavily in security. That's not just someone getting access to your account directly, but your experience of someone else's account getting compromised. If your friend gets compromised, you feel it. It undermines your trust in your experience.
The network effort can be used for bad as well as good. That's why we've invested heavily in security for a long time. No individual can be on top of all the different risks every day-it has to be orchestrated across a bunch of different groups.
I'm always concerned about the risk of a compromised account. There are high-profile individuals, companies and governments that use Facebook as a way to communicate. That means we need to make sure they're comfortable coming on to Facebook and feel secure in using it. We've seen situations where high-profile accounts get compromised. That is guaranteed to draw attention and undermine trust.
CSO: What is your strategy for dealing with misappropriated credentials?
Sullivan: We created some great technology modeling the behavior of a real account. It's machine learning. We have a large group of engineers working on this all the time.
Six hundred thousand times a day, someone tries to log into account using [stolen login information]. We catch them and block them.
I had a meeting with someone from a vendor earlier today, and he told me he tried to log into his Facebook account on the hotel computer [and was presented with a security challenge question]. He was coming from a public computer, and he had the right password. [But] public computer plus different state, that triggered social verification process. Some services would just look at that type of activity as a risky log-in. We do something different, which is social verification. We presented him with profile photos of his friends [and had him pick their names].
We've gotten better at this. Using machine learning, we can figure out who are the friends you interact with the most on Facebook.
CSO: How is this learning accomplished?
Sullivan: If I [refuse a friend request], I have now sent two different messages to Facebook about that account. Even in a single friend request, we can see a lot of different ways to learn from that experience. We call it the Facebook immune system. It's basically scoring every interaction against the site.
Looking [back at the earlier example], what if he was coming from [his home] state but was sending 100 messages per hour, as opposed to last month when he sent 100 messages in all? We can detect that change in behavior. Credit card companies do [something similar.] One time I flew to San Francisco for work, and I had some spare time so I bought my wife an engagement ring. (I didn't buy her a ring when we got engaged.) So, in San Francisco I used a credit card to buy a ring for her. They called her and got her to validate that it was an acceptable purchase. With the credit card industry, they're customizing their flags to their environment and the patterns of abuse they see.
When you show up on day one as a new user, you start acting a particular way. If you show up as a fake user, you start acting a different way. So, the best security feature on Facebook is something we don't have to do; it's the reporting mechanism we provide for the people who use Facebook. It's not just that you are a fake user and you send an inordinate number of friend requests to a category of users. You actually also set off alarms to other people.
A fake account thrashes around in the Facebook environment so differently from the way a real person behaves. It's like the largest and most effective Community Watch program in the world.
CSO: What are some other trust issues you've dealt with?
Sullivan: A couple of years ago, the biggest bane of Facebook was half the people loved to use games, and the other half hated hearing about the games. So, we had to respond to that, but we also had to address the underlying issue, which was that [people] didn't want to spam their friends. For the game developers, the more you spammed your friends, the more successful they were.
Now, if you want to play games on Facebook, you can have that experience without sharing what you're doing. If you don't want anything to do with Words With Friends, for example, you have the ability to hide anything having to do with Words With Friends. If my news feed shows one of my friends is listening to a song on Spotify, I can hide it, report the story as spam, or I can dial up or down how much I hear from this friend. Or there is another option, "Hide all by Spotify." You can make it go away.
Next page: Fake accounts, Google, Twitter, and Timeline...