What does Amazon's S3 Server-side encryption protect against?

Question

Amazon's S3 storage service offers server-side encryption of objects, automatically managed for the user (Amazon's Documentation). It's easy to enable so I'm thinking "why not?", but what kind of security does this really provide?

I guess it prevents someone from wandering into the AWS datacenter and grabbing a hard drive, but that seems very unlikely, and presumably anyone with access like that could also get the AES keys, wherever they're stored.

It doesn't seem to protect the data once it's off the drives, since at that point it's decrypted, and anyone who has your credentials or can intercept the traffic will see the data in the clear. So what's the point, really? Just to say the data is "encrypted"?

Answer 1

The short answer is this: We have no idea, probably none.

It might protect against stolen backups. But that assumes Amazon even makes backups. That seems very unlikely. If they did, why couldn't they recover data from their last S3 data loss? It's much cheaper and more efficient just to use multiple live copies.

Also, Amazon would need the keys on every access. So it seems very unlikely that they store the keys anywhere other than approximately the same places they store the data. So if you're imagining a theft of live data devices, it's just as likely that they get the keys as well.

But we don't know how Amazon stores, replicates, and/or backs up data. Nor do we know where they store the keys or how they distribute them. However, I've yet to hear a plausible argument that there exists a realistic threat they protect against. The "stolen backups" theory seems to be based on the false premise that Amazon uses backups when all the evidence suggests they use multiple, live copies with the keys quite nearby.

Dropbox's encryption, however, does protect against one real threat model, albeit a very unlikely one. Dropbox stores their own keys and sends them to you, so it does protect you from a rogue Amazon employee. In exchange, you're vulnerable to a rogue Dropbox employee or Dropbox security bug.

My own opinion is that Amazon added this feature just so they could say that data could be stored encrypted. Some people will mindlessly compare check boxes on feature lists and Amazon wanted a check box on the "secure/encrypted" line. Either way, the weakest link is most likely Amazon's internal network and human security and the validity of the implementation of the code that decides whether to permit accesses or not.

Answer 2

I guess it prevents someone from wandering into the AWS datacenter and grabbing a hard drive, but that seems very unlikely, and presumably anyone with access like that could also get the AES keys, wherever they're stored.

Gilles' comment effectively answers your question, really, but I'll go with a longer answer myself because I'm nice. Disk encryption protects you against data loss when a disk is stolen and the key is not stolen with it. Such examples might be, as Gilles says, stolen backups, but could also be in laptops on the move, or disposed of hard disks to prevent meaningful attempts at salvaging data from your decommissioned disks.

Disk encryption doesn't do much to help you when you put the key and the disk together, because the security relies on the key and if the key can be intercepted, the data can be decrypted. The key and the disk are always in close proximity by necessity when the OS is on and using the disk (every read requires that key) so anyone near it who can reasonably intercept the key should be able to read the data. Of course, you do need to be able to recover the key to effect any kind of attack, so it is slightly harder than just copying a hard disk (but not by much). So basically, yes, you're right.

However, it is still a good idea to protect your disks to minimise the potential loss of data through things like theft and disk disposal. You don't know what or how Amazon do to destroy those disks, so if you have valuable information on there of any kind, having them encrypted is a great idea.

So what's the point, really? Just to say the data is "encrypted"?

That is actually a possible factor. As I say there are tangible benefits from encrypting data are not quite those you might expect, but still exist. That said, I have had customer requirements that data be encrypted on the server end in a similar scenario as a marketing point (we encrypt your data). I think there's an educational challenge there for security people.

Answer 3

You can also be protected against someone breaking into the disks at S3 - say (humorously) that the disk that the S3 data was stored on is also a boot drive for a working windows XP box, and someone breaks into the XP machine (not physically - through a hack). They then have all the files on the machine, but yours are encrypted with keys stored on some other box, so all the thieves get is digital rubbish.

Perhaps the likelihood of someone breaking into an S3 array is small, but I side with other posters and bet that the keys are behind another wall. Also they likely use different keys for each account.

So while its not a ton of security, it is there. Dropbox has one giant de-duplicated map for its store, so I don't see how they could encrypt with different keys for each account.

Answer 4

Some things to remember:

• Amazon is used by a vast number of companies
• Lots of valuable data in there: financial data, intellectual property etc
• Criminals like targets like this, that can return high cash value
• Crime groups are not averse to placing individuals within datacentres, or forcing employees to carry out nefarious tasks

Don't overlook the problem of your data being leaked, deliberately or otherwise, by third parties, even ones as big as Amazon.