One of my co-workers had some questions about email encryption and how it worked, so I ended up writing him a long response that I think deserves a wider audience. Here’s most of it (leaving out the NDA covered portions).
Email Encryption and HIPAA Compliance for the Uninitiated
In IT security, when we talk about encryption, there are a couple of different “types” of encryption that we worry about, one is encryption “in transit”, and the other is encryption “at rest.”
Encryption “in transit” is how we ensure that when data is moving from one system to another that it is either impossible or difficult beyond reasonable likelihood for someone to intercept and read that data. There are pieces of many data exchanges that we have no control over, so we cannot guarantee that there isn’t someone out there with a packet sniffer reading every bit that passes between our server and someone else’s (This is a form of “passive” data inspection, possible from just about any trunk line on a switch). We can make sure it doesn’t happen on our end, but we can’t control the ISP or the other person’s side of things.
The basic email encryption system, TLS (Transport Layer Security…Don’t ask what that means), usually follows this incredibly oversimplified pattern:
1. Server 1 contacts Server 2
2. Server 2 says, “Hi. I’m Server 2. Who are you?”
3. Server 1 says, “Hi. I’m Server 1.”
4. Server 2 says, “Nice to meet you Server 1. What can I do for you?”
5. Server 1 says, “Before we really get into that, I’d like to make sure no one is eavesdropping on our conversation. Can we start talking in a language no one but us knows?” (This is basically what encryption is)
6. Server 2 says, “Sure. What language would you like to use?”
7. Server 1 hands server two a certificate that serves as a kind of translator, which Server 2 will use to translate (decrypt) everything that Server 1 says from now on. Server 2 will also use this certificate to send any responses or other messages back to Server 1.
8. Server 2 says, after translating what it wants to say into the new encryption language, “Okay, what would you like to do?”
9. Server 1 translates this message from the encrypted language and makes its first request to server 2 after translating it into the encrypted language.
From this point on, each server will communicate exclusively with the encryption “language” provided by the certificate they exchanged, and anyone who is eavesdropping (packet sniffing) will only see a bunch of gobbledygook that they can’t understand.
There are more complex versions of this scenario that make things more secure. For instance, in a Domain Authenticated TLS situation, both servers have to be “Authenticated,” which is to say, they must prove they are the server the message is supposed to go to. This is done by validating the name that is printed on the certificate with the name the servers use when introducing themselves to one another.
In the example above, it is possible for someone to inject themselves into the conversation and decrypt everything from server 1, read it, encrypt it again, and send it on to server 2 (this is called a Man-in-the-middle attack, and is an “active” form of eavesdropping, because it requires a very complex setup and specialized hardware to accomplish, and also requires active manipulation of data that is being inspected). Domain Authenticated TLS makes this much more difficult, because a server that acts as a mediary in a man-in-the-middle attack cannot use the name that exists on the certificate unless it is owned by the entity that created the certificate to begin with. When you get certificate errors while browsing the web, this is usually due to either you entering a name that isn’t listed on the certificate that is installed on the server you’re talking to, or the server is using a name that isn’t listed on the certificate. (Certificates are a heavy subject, so I’ll just bypass that for now)
Anyway, data “at rest” is any data that is just sitting on a hard drive or disk somewhere, waiting for someone to read it. In order to read that data, you have to gain access to a server (or workstation) that has access to the data and read it. Encryption of data “at rest” requires more effort to accomplish, because it has to be decrypted every time someone tries to read it. Technologies like Bitlocker or PGP allow data to be encrypted while it’s just sitting there on a server.
We only care about encryption of data “in transit” when we work with HIPAA regulations. This is because the only way to access data that is “at rest” is to gain physical access to the data or to systems that have access to that data. HIPAA has other regulations that help reduce the likelihood that either of those things will happen, and since data “at rest” is never outside our realm of control, we can do much more to protect it. Most ePHI is sitting in a datacenter that is locked and requires specific permission to access, but that coverage doesn’t apply to the data when it’s moving between servers.