I actually tested that a few months ago. Every block was fully binary zero.
Volumes are logically allocated in a manner similar to how files are allocated in a filesystem. When you create a file in a modern filesystem, it NEVER sets up data block pointers to existing blocks that have data. You cannot see latent data by creating an empty or sparse file and reading it. The only way a file's logical location can end up with a pointer to a disk block is by actually writing it. If that's a partial write, it will TRY to read the block first so that partial data is retained. However, since this is the first use of that block, no pointer exists, so the OS just fabricates a block of binary zero content, updates with the partial write, writes it to disk, sets up a pointer, and writes the block that contains the pointers.
Volumes work much like a file in a massively huge filesystem, but are simply accessed by the mechanism AWS has deployed (I guess a very scalable, reliable, and redudant SAN/NAS system). I do know some details about one SAN system and it works like that.
Your data is safe at AWS to the extent you can trust that staff at Amazon to not mess with things. Most staff would have no level of access that could do it. The few that do would likely have undergone some serious background checks (as I have gone through just to be a sysadmin at a company that is an AWS customer). Furthermore, the amount of data and processes going on at AWS would just boggle your mind. That'll keep staff busy.
My ONLY concern at AWS is a concern that would be present regardless of the provider ... that the government of the jurisdiction they are in may possibly be able to demand the data. Look up "megaupload". But AWS at least has a choice of jurisdictions.
