The best defense is being someone too uninteresting to bother. Once you're interesting so some resourceful adversary, it's very hard to avoid devices being hacked, and virtually impossible to determine if they've been hacked.
So that's not a particularly viable option.
(Difficult to prove, natch, but possible.)
We know that the NSA intercepts machines for modification. And it's possible that hardware is generally backdoored. Maybe even by Chinese manufacturers.
But what can one do, if everything is pwned? It's not practical to build machines from transistors etc. There are dreams of open-source hardware. But how could that even be done securely? The NSA can plant agents anywhere, in theory.
I'm increasingly of the view that it's not, at least not through individual action.
My interest is, for first strokes, painting an accurate picture of the landscape. Which means discarding inaccurate models and frames.
Among those: that laying low is possible, or a positive (that's precisely the objective of the Panopticon, and self-censorship abd -reegulation are the most efficient), or that individual rather than collective action is appropriate.
It also seems that surveillance itself faces various realities and economies, which can be directly attacked.
A much smarter approach in my opinion is to live on two separate layers: a secure private underground and an uninteresting public surface.
This not only keeps the enemy content, it also keeps tracking low; any open confrontation will necessarily lead to harsher measures, which ultimately means violence.
If samizdat worked in the URSS it can work nowadays too.
Yes. And never connect layers. Mirimir has no meatspace contacts. Also, one can have more than two layers.
This is what I call "defense by presumed motive" and is flawed.