How to ensure data protection on remote server?

Question

We have sensitive data from a customer and need to train a Neural Network for our customers.

However we have no severs to train the Neural Networks, so we want to rent GPU servers. There are many cloud services with affordable prices and we can get root access to them.

The problem is, that we aren't guaranteed that our data is safe on those servers, which is very important for our customer. We want to ensure that the server's hosting party has no also root access and could not theoretically copy our code and data.

Is there a way to run our Python code safe ?

We could make a binary of our Python program but it could be decompiled in data and code, right?

We could also install an virtual machine OS inside, but is that safe for our data ?

Is there a docker image for an encrypted OS , or does something like this exists ?

Is there a solution for our problem out there except of buying immense hardware ?

Thanks a lot for bringing in your experience and helping us!

Answer 1

First of all, all data you can compile can be decompiled if you really want its a question about time, so you are correct here.

I did not know where in the world you are, but I will recommend AWS, Azure, Google Cloud and some of the big players out there, its not cheap but here you are sure about the provider did not have your root access key.

If you lose your root key or SSH key you its will require you can create a snapshot of the VM disk and restore it for a new VM.

Both AWS, Azure and Google have a very high level of security and use a vary advance ACL for users so its your responsibility to have the security ofc.

I'm located in EU and working for a company where GDPR is very important, so I will recommend one of the big players, and read the data protection privacy. If needed contact a lawyer to review the company based on your use case.

If its still not securer enough for the data you working with, I will highly recommend locally hardware and here the customer shut pay for that hardware fee you spent based on your contract.

that's my recommend