The real solution is high assurance systems: those designed in a rigorous way with an argument that they'll remain secure against any known classes of attack. That's every layer of the stack and sideways as my proprietary framework [1] used to show. Here's an example [2] of what that's like applied to Tor. Ottela's Tinfoil Chat is only one I know that applies high assurance at the design level to eliminate risk of High Strength Attackers. He took a lot of feedback from us on Schneier's blog and applied it well.

Client-side encryption using C libraries on Linux or whatever is just asking to be hacked. Best will be secure hardware, drivers, kernel, and trusted components. See CHERI processor [4], EROS documentation [5] (esp "Principle-driven...") and Mikro-SINA VPN [6] for examples of The Right Thing in security-critical development. Need more like that.

[1] http://pastebin.com/y3PufJ0V

[2] https://www.schneier.com/blog/archives/2014/09/identifying_d...

[3] https://github.com/maqp/tfc-otp

[4] https://www.cl.cam.ac.uk/research/security/ctsrd/cheri/

[5] http://eros-os.org/eros.html

[6] http://genode-labs.com/publications/mikro-sina-2005.pdf

You can still learn a lot about a social network just by analyzing traffic patterns and meta data. Encryption by itself is not a panacea.

Client side encryption, and no physical access to the outside... and...

Considering that many encryption systems around seem to get broken, I don't think that's the only solution needed.