[talk] a Guardian article

Antti Kantee pooka at iki.fi
Fri Mar 31 12:32:39 EDT 2017

On 31/03/17 15:56, George Rosamond wrote:
> Ike had a story about Russian mitigation of German radio surveillance
> during WWII. The Russians had numbers and relied on that for their
> solution. *Everyone* spoke on the wire, about *everything*. The weather,
> the latest radio broadcast performances, their kids' birthdays... and
> since German radio surveillance relied upon individuals who could
> understand Russian, the volume was beyond their capacity. The German
> surveillance solution couldn't scale. Meanwhile, unencrypted high-value
> communications continued in this large pool of noise.
> Apparently this tactic continued well into the Cold War against US
> surveillance.
> The difference today is scaling. Digitized communications are easier to
> collect in volume, and to store and process, without relying on
> individuals to listen-in.

Nice story.  I assume they were talking about the weather etc. because 
everyone is not capable of creating believable military-like traffic. 
Since everyone can listen now, like we've been discussing, the key is to 
figuring out how to generate military-like traffic instead of just 
birthdays and the rainy weather.

> Digital surveillance can also more easily parse and discover anomalies
> in  the traffic. And then it can be correlated with cell phone traffic, etc.

That's just one more reason why my cell phones only do voice/sms, and 
why I don't really carry them around all that often.  (well, the real 
reason is that I don't get disturbed very often that way)

> Your local ISP surveillance device says: "I am watching lots of traffic,
> and while I see the usual queries to the news www sites, I also see that
> they are accessing Chase bank on the first of each month."

That's the computer science solution to figuring out which bank we use. 
The [computer] engineering solution would be for the ISP to look at the 

> The point is white noise needs to be customized insofar as anomalies
> can't be easily identified.  It has to be lots of banks that are
> accessed, including (you) the target's own banks, at regular sloppy
> intervals.

Maybe the solution is to feed normal traffic into some machine learning 
algorithm?  I've been wanting to do machine learning anyway, here's a 
nice itch to [not] scratch.

More information about the talk mailing list