They cluster their databases across multiple instances, availability zones and regions, and back them up constantly to S3. Their ultimate recovery plan is to restore from backup if necessary, which is pretty much the same as assuming EBS will corrupt your data.
While it's far from trivial to do it that way, they seem to be the most successful user of the AWS systems.