Big Storage
If we want security in the context of big data then we must trust modern big storage.
There are ways to organize data in big storage so as to detect storage failure, either natural or malicious.
Merkle trees of data can track retrieval errors.
I am now thinking about those ideas.
They apply perhaps over several orders of magnitude; from 106 bytes on up.
They add a complexity burden.
High statistical reliability is also available with Reed-Soloman coding, at an additional complexity burden.
One plan is to keep the entire branch of the tree in trusted memory and update the tree as any storage, including the hashes, to untrusted memory.