It's a bit disturbing to me that they seem to be using AWS for confidential editorial work.

> Due to editorial requirements, we needed to run the database cluster and OpsManager on our own infrastructure in AWS rather than using Mongo’s managed database offering.

In a happy world the guardian wouldn't rely on a company we spend a lot of time reporting on for unethical practices (tax avoidance, worker exploitation etc.) - but we decided it was the only way to compete. One of the big drivers was a denial of service attack on our datacentre in 2014 on boxing day - not an experience any of us want to have to deal with again.

>Since all our other services are running in AWS, the obvious choice was DynamoDB – Amazon’s NoSQL database offering. Unfortunately at the time Dynamo didn’t support encryption at rest. After waiting around nine months for this feature to be added, we ended up giving up and looking for something else, ultimately choosing to use Postgres on AWS RDS.

Anyone who gets control of the live server can still read a database, even if it encrypts its storage.

Exactly. As I read the original article, which mentions "encryption-at-rest", there was a voice in my head crying: "No, what they need is E2EE". That would enable the authors to write confidential drafts of the articles, no matter where the data is stored (and AWS would be perfectly fine of course).

Disclaimer: The voice is my head does not come out of nowhere. I am building a product which addresses this: https://github.com/wallix/datapeps-sdk-js is a API/SDK solution for E2EE. Sample app integration is available at: https://github.com/wallix/notes (you can switch between master and datapeps branches to see the changes of the E2EE integration)

In which case they could've just used a separate encryption layer with any database, including DynamoDB. The HSM security keys available from all the clouds makes this rather simple.

Yes, any db including Dynamo would have been fine.

Our software E2EE solution has advantages over HSM though: Cost obviously, and more features and extensibility.

Great idea.

Encryption at rest is still important as it closes off a few attack/loss vectors: mis-disposed hard drives, re-allocated hosts. I'm probably missing a few others.

Yeah, but it doesn't really address my concern.

Anyone who can control a server in any environment can potentially interact with the database powering applications running on that server.

How is running on AWS different than Guardian Cloud in their basement?

The level of control over who has physical access, of course.

Did reports of the Snowden revelations reside on the CMS?

Sadly we don't trust our security practices anywhere enough for that! Secret investigations happen in an air gapped room on computers with their network cards removed then get moved across to the main CMS when they're ready to publish.

Probably not, no, until they were about to be published. I imagine that the choice between "run an entire data centre ourselves, store everything there" and "use AWS, but keep high sensitivity stories on local machines" is an easy one.

After all, the client computer that connects to the CMS is just as, or more likely to be compromised. I wouldn't be surprised if the coverage (or at least parts of it) were edited on airgapped laptops.

> the choice between "run an entire data centre ourselves, store everything there"

If those were the only two choices, you might be right. But the resources needed for the actual CMS functionality sound modest enough to run independently of the main website.

> the client computer that connects to the CMS is just as, or more likely to be compromised

That's faulty reasoning.

> That's faulty reasoning.

Why? It's an obvious potential point of compromise.

Sorry, I misunderstood. I read it as saying "We're going to get hacked via this other vector, anyway, so why bother?" I see your point, now.

They're using AWS VPC (Virtual Private Cloud) which isn't open to the world (you use a VPN to bridge the VPC into your internal network) and which you can spin up dedicated instances that don't share underlying hardware with other AWS customers.