The system is going to fall. The ultimate example of planned obsolescence. Whether Assange is completely successful in his endeavors or not. Whether he is jailed or not. It was designed that way.

Assange certainly is not a libertarian (or any faction of the US political diaspora). I think you are dead wrong on his understanding of complexity. More to the point, it appears Assange is counting on it. When dealing with something of this complexity, you've got to think of it more like reinforcement training in a neural network. Put roughly, the theory being that society can only operate as a healthy network in making good decisions if the agent performing action y[t] which forms the basis of observation [x]t and assessment of instantaneous cost [c]t is actually performing the actions they say they are. Otherwise x[t] and c[t] are weighted based on an invalid set of assumed actions leading the network to associate the observed outcome x[t] and assessed cost c[t] to the wrong action y[t] ... system weighting either in the direction of or away from that particular y[t] in a fashion unhinged from reality.

It gets even more complex because what we really have is the agents who serve the role of feeding y[t] into the public neural process intentionally providing an invalid description of the actions - or trying to keep actions from becoming training inputs at all. These folks are maintaining their own parallel structure trained with accurate inputs for y[t] (from the standpoint of their own actions) with expected outcomes from those using the "public neural network" intentionally trained with inaccurate data becoming an additional y[t] point in their own neural process.

What Assange is doing is weighting nodes - regiggering costs like crazy. For example, from the NYT perspective there is now a whole different cost assigned for parroting a "fact" on the wars or diplomacy from "unnamed administration officials" which everyone knows is untrue yet niceties require "journalists" to pretend otherwise. If something they print can be disproved with primary source materials available to anyone, publishing inaccuracy becomes a question of credibility for the organization ... if discredited, those who use it as a mouthpiece certainly wouldn't find the Grey Lady of such utility, so they tend to be more weighted to not using the tactic of asking the press to print untrue stuff anonymously. It's not just one outcome, it's tuning the system to statistically be weighted in one direction vs. another.

There is no way in hell to just shut the machine down nor is that even necessarily a desired outcome (I think that would ultimately require extinction). It's more a matter of shorting out a series of bad decision-node clusters. As you point out, the function the network will be solving is always going to be "self preservation". His theory is that if enough nodes that are currently low-cost due to secrecy are weighted more heavily, the expected cumulative cost will tend to push solutions away from those nodes. The goal is to weight things in such a such a way that the least-cost solution to self preservation goes through a more desirable path. He believes that this is most likely if there is an accurate public perception of y[t] as it relates to x[t] and c[t] - if the neural processes used by both the public and policy makers are generally tuned. This idea actually relies on society behaving in reasonably close adherence to complexity theory in order to work.

I'm not saying his theory is 100% right, time will tell, but don't sell it short. They've only been provided leaked documents related to the US within like the last 10 months or so, but Wikileaks has been doing what it's doing for quite some time. Too many people are trying to make Wikileaks out to be a group that exists to oppose or embarrass the US. Until quite recently our politicos were celebrating them getting information out from some seriously repressive regimes.