Design Challenge: Sybil Attacks

If it is possible for anyone to connect to a computer system then it may be possible to interfere with the operation of that system, and defenses against interference must be designed in.

For example, email's design is vulnerable to spam. To send someone an email, all that is required is to have their email address; email is unsolicited messaging, so it's easy to send nuisance or fraud emails.

Spam Filters

Email systems deal with spam by filtering it out. There are two basic approaches: The simplest is pattern matching – messages are compared to models of the sort of messages that spammers send. This produces an arms race between spammers and spam filterers. Spammers must strive to create messages which beat the filter, while filterers have to improve their model.

One particularly interesting example of this is (snail) mail advertisers that employ handwriting services to write the addresses on their envelopes. This fools the receiver into believing that an actual human sent them the letter making them much more likely to open it.

Social Networks

The other approach is to use social/trust networks. Services like Facebook or Twitter are significantly less vulnerable to spam (if you don't consider the ads spam). Instead of receiving personal messages from anyone one who knows your id/address on a social network, you only receive messages from other users you have "followed" or "friended".

We like to call this a "Solicited Spam" system. Following someone on Twitter means you think they are human, or at least, an amusing bot. This is actually a pretty strong signal. Email doesn't have an explicit social network, but the best spam filtered email systems (like gmail) can use the implicit social network in email: if I send you and email and you reply then we are "friends".

Then graph analysis can be applied to the friend network – spammers may be isolated, or clustered together and can be filtered out.

By building SecureScuttlebutt around a social network, preventing spam becomes easy. Since, in a social network application, you only communicate with trusted peers (friends) anyway, and there is usually an explicit "follow" mechanism which expresses who you trust, and humans are pretty good at detecting frauds. This is why SSB was made as an inherently social system.